Files
pbakaus_impeccable/tests/live-e2e.test.mjs
T
e2ff625b63 The Rust engine: one binary replaces every script and the JS detector, fully open (#714)
* Add oracle harness: verb goldens and function-level vectors

Records stdout/stderr/exit/files for every impeccable verb over a fixed
corpus and replays them against an alternate implementation. Adds a loader
hook that captures per-function call vectors from the pure engine modules.

Prepared with AI assistance (Claude Code).

* Oracle: hook, hook-before-edit, hook-admin cases and goldens

Prepared with AI assistance (Claude Code).

* Add docs/CLI-CONTRACT.md: observable behavior of every impeccable verb

Prepared with AI assistance (Claude Code).

* Oracle: context/doctor/pin/surface-brief/critique/palette/embed/signals/csp/seed/genimg/question cases and goldens

Prepared with AI assistance (Claude Code).

* Oracle: live-mode cases and goldens (roots, inject, wrap, insert, accept, session, manual edits, daemon)

Prepared with AI assistance (Claude Code).

* Oracle: mask the binary path before HOME; export launcher env to the binary

Prepared with AI assistance (Claude Code).

* detect: set process.exitCode instead of exiting after the final write

process.exit() right after a large piped stdout write truncated JSON output
at the pipe buffer boundary; found by the oracle harness. Re-record the six
directory-scan goldens that had captured the truncation.

Prepared with AI assistance (Claude Code).

* Oracle: normalize the hook-admin command in both runtimes' forms and audit chars

Prepared with AI assistance (Claude Code).

* Skill text: invoke the impeccable launcher instead of node scripts

Every `node {{scripts_path}}/<name>.mjs` becomes `{{scripts_path}}/impeccable <verb>`
(context-signals -> signals, hook-admin -> hooks). Setup step 1 drops Node, points
Windows shells without sh at impeccable.cmd, and says the launcher runs a
self-contained binary. allowed-tools follows.

Prepared with AI assistance (Claude Code).

* Scripts dir: replace the Node scripts with the impeccable launcher

skill/scripts keeps command-metadata.json and the page JS; every .mjs entry
point, lib/, and live/ are gone (the binary owns those verbs). Adds the POSIX
launcher, impeccable.cmd, VERSION (copied from the new root ENGINE_VERSION),
scripts/fetch-engine.mjs (bun run fetch:engine) to pull the pinned binary
into skill/scripts/bin/<os>-<arch>/, and gitignores that bin dir.

Prepared with AI assistance (Claude Code).

* Build: ship the launcher instead of bundling the JS engine

readSourceFiles no longer copies cli/engine into the skill; the scripts
payload is the launcher (executable bit preserved through dist, plugin/, and
universal.zip), impeccable.cmd, VERSION (synced from ENGINE_VERSION on every
build), the page JS, and command-metadata.json. Hook manifests call
`<scripts>/impeccable hook` behind an existence guard (Codex adds a
commandWindows sibling calling impeccable.cmd; Cursor runs hook-before-edit;
GitHub keeps the git rev-parse form; Grok mirrors Claude); the Node probe and
systemMessage notice are gone. build:release fetches the pinned engine for
every target (lenient) and stages bin/<os-arch>/ into the dist skill copies
after root harness dirs and plugin/ were synced, so git-delivered trees stay
launcher-only. The detection-rule count check reads the vendored
extension/detector/antipatterns.json and is skipped when absent.
build:browser is a stub; the codex prefix rewrite leaves
`{{scripts_path}}/impeccable` alone.

Prepared with AI assistance (Claude Code).

* CLI: turn the impeccable npm package into a platform-binary shim

cli/engine, cli/lib, and cli/bin/commands are gone; their behavior lives in
the engine binary. cli/bin/cli.js now resolves the binary from IMPECCABLE_BIN,
the @impeccable/cli-<os>-<arch> optional dependency (templates under
cli/platform-packages/, published by the engine release), the
~/.impeccable/bin/<version>/ cache, or a checksum-verified download, and
execs it. package.json drops the engine dependencies and the library
exports; puppeteer moves to devDependencies for the icon scripts.
README.npm.md describes the shim.

Prepared with AI assistance (Claude Code).

* Tests: gate behavior on the oracle and the engine binary

Unit tests of the deleted Node scripts and the JS detector are removed;
their behavior is pinned by tests/oracle goldens (frozen JS behavior plus
reviewed deltas) and the engine's own tests. tests/oracle.test.mjs replays
the corpus against the binary (IMPECCABLE_BIN or skill/scripts/bin/<target>/,
via tests/lib/engine-bin.mjs) and skips cleanly without one; the framework
fixture sweep drives live-inject, live-wrap, and detect-csp through the
binary the same way. record.mjs learns --bin. The function-level vectors
under tests/oracle/vectors/calls are committed as the frozen snapshot they
can no longer be regenerated from. Suites: core trimmed to build and
transformer tests, oracle added to the default run, detector/live reduced to
packaging and reference checks, the live-e2e helper tests move to the opt-in
live-e2e lane pending its retarget, cli-remote-e2e is an empty placeholder.

Prepared with AI assistance (Claude Code).

* Docs: describe the launcher, the engine pin, and the oracle gate

CLAUDE.md gains an Engine binary section (launcher lookup order, ENGINE_VERSION,
untracked binaries, how tests get one, the oracle as behavior gate, what stays
JavaScript) and drops the Node-script and JS-detector descriptions; the CLI
and detection-rule sections point at the shim and the engine repo. README.md
states the skill needs no runtime and lists the launcher-based hook commands;
AGENTS.md follows. CLI-CONTRACT.md's intro notes the scripts it quotes are
the recorded source, not the tree.

Prepared with AI assistance (Claude Code).

* Tests: tighten the hook command guard assertion

Prepared with AI assistance (Claude Code).

* Oracle: re-golden 46 cases for the engine's own command names; record them in DELTAS.md

Prepared with AI assistance (Claude Code).

* Build: ship launcher-only release zips by default

IMPECCABLE_BUNDLE_ENGINE=1 opts in to staging the engine binaries into the
dist skill copies. Bundling every target into every provider copy put
dist/universal.zip near 340 MB, past the 25 MB Cloudflare Pages file cap
that impeccable install downloads through.

Prepared with AI assistance (Claude Code).

* Tests: drive the live-e2e orchestrator through the engine binary

The session, fake-agent loop, steer test, and manual-edit probe spawn
<binary> <verb> (live-server, live, live-inject, live-wrap, live-insert,
live-accept, live-poll, live-complete) resolved by tests/lib/engine-bin.mjs
instead of node skill/scripts/live-*.mjs; the completion typing the agent
imported from the deleted live/completion.mjs is a small local helper. The
live-e2e helper unit tests move back into the default live suite (the steer
loop skips without a binary).

Prepared with AI assistance (Claude Code).

* Tests: run new-work-e2e through the engine's serve-question and generate-image verbs

Prepared with AI assistance (Claude Code).

* Tests: point the skill-behavior harness at the launcher and engine binary

The bash tool exports IMPECCABLE_BIN so the staged skill's launcher runs
without a download; scenarios assert on 'impeccable context' instead of
context.mjs and skip without a binary.

Prepared with AI assistance (Claude Code).

* Tests: note what plugin-e2e validates before and after the generated-output sync

Prepared with AI assistance (Claude Code).

* Oracle: record the engine's 'wasm-unsafe-eval' CSP meta patch as a reviewed delta

Prepared with AI assistance (Claude Code).

* Rebase reconciliation: fold main's post-freeze work into the swapped tree

The rebase onto origin/main brought changes whose JS engine halves left the
tree with the swap. This commit reconciles what survives:

- Suite map: register main's comp-fidelity unit tests (build-phase,
  comp-diff, font-match, hero-checks) in the core suite and
  live-browser-ignores in the live suite.
- Payload guard: the skill scripts payload now allowlists the comp-fidelity
  build pipeline (comp-spec/comp-diff/build-phase/font-match and their libs),
  the one Node toolchain that has not moved into the engine.
- Drop skill/scripts/live/project-ignores.mjs, lib/live-path-globs.mjs, and
  their test: they import hook-lib/live-inject/impeccable-paths, which the
  swap deleted, and their consumer (the JS live server) is the engine now.
- skill text: the comp pipeline's calls to engine verbs (generate-image,
  embed-prompt) use the launcher spelling.
- Oracle: re-record 17 detect goldens over the fixture set main changed
  (oklch #592, color-mix #578, 1D grid #615, the two comp-fidelity rules)
  and record the gap in DELTAS.md; those JS rule changes are not yet ported
  to the engine, and the goldens pin its current behavior.

bun run test (oracle included) and bun run build are green on this tree.

AI-assisted change: implemented with Claude Code.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WaJv2c4oN8wS7Ttq4XRqyx

* Launcher: engine-probe PATH validation, working .cmd download path; CI: drop stale path, add oracle job

Byte-identical copies of the engine repo's launchers (engine main
af7572c): the retired 3.x npm CLI on PATH or in ~/.impeccable/bin is
rejected by the engine-probe handshake instead of hijacking every verb;
impeccable.cmd's download path is rewritten as straight-line goto flow
(the parenthesized blocks expanded %url%/%cached% at parse time, making
it dead code) with certutil sha256 verification and a windows-arm64 ->
x64 asset fallback; the final error points at the release download
instead of npm i -g (npm still serves the 3.x CLI).

ci.yml: the generated-output check no longer diffs the deleted
cli/engine/detect-antipatterns-browser.js, and a new oracle job fetches
the pinned engine (bun run fetch:engine) and replays tests/oracle/
against it. The job is continue-on-error with a loud warning until the
first engine release exists; flipping it to required is a release-time
toggle, documented in the workflow.

Verified here: sh -n on both launcher copies, bun run build green, full
oracle replay against the rebuilt engine binary green (770 pass, 0
fail), and a launcher behavior test proving a fake 3.x CLI on PATH is
skipped while the download + checksum chain completes against a local
file server.

Prepared with AI assistance (Claude Code).

* Oracle: restore detector goldens to post-fix behavior after the engine ports

The Aug 17-31 detector fixes (oklch parsing, color-mix nested hex, 1D grid
pass, comment stripping, root-relative linked stylesheets, URL userinfo
redaction, inert ignore-value refusal) and the comp-fidelity rules
organic-clip-path / buried-raster are ported to the engine. Re-records the
gap-pinning detect goldens from the fixed binary (glow.html included: its
.photo-opaque-grad column now carries the buried-raster finding it was
written for), replays the frozen checkHtmlPatterns call vectors through the
last JS engine state in history (db1462b9^; args untouched, 14 of 101
results moved), and rewrites the DELTAS gap section into the landed-ports
note. Each re-recorded json fixture golden byte-matches that JS state's
output; oracle: 770 pass, 0 fail.

Prepared with AI assistance (Claude Code).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WaJv2c4oN8wS7Ttq4XRqyx

* Oracle: pin the Aug 17-31 verb fixes ported to the Rust engine

New cases: hook-session-grok-edit-then-stop (Grok Build camelCase envelope,
end_turn/shutdown/stopHookActive Stop handling, 35ae0733 + bfe634e2 +
3c442af7, #646), hook-session-codex-stop-decision (Codex Stop emits
decision/block, c9e7cd8a, #603), and doctor-order-boot-and-deep (boot and
deep findings keep their established artifact order, 80997663).

Re-recorded goldens whose old bytes froze pre-fix behavior, with a
DELTAS.md entry naming each upstream hash: the Stop finding-cache sync
(3c442af7), the Edit|Write manifests without the retired MultiEdit matcher
(7d5c60d2), and the failWithRollback field order (1f2c3f9d).

Prepared with AI assistance (Claude Code).

* Oracle: drop a duplicated DELTAS section

The verb-fix section landed twice when two porting sessions staged the
same file; keep one copy.

Prepared with AI assistance (Claude Code).

* Oracle: pin the hooks ignore-value inert-entry refusal

Three hadmin-ignore-value-inert-* cases record the engine's port of
be87f5eb (#662) to hooks ignore-value: an exact value for a rule whose
findings can never extract one is refused with the wildcard-plus-file
route (and no config write), while the wildcard scoped form for the same
rule is accepted. Goldens recorded from the engine binary and verified
byte-for-byte against the ea360025 hook-admin.mjs on the same sequences.
No existing golden changes, so no DELTAS entry is owed.

Prepared with AI assistance (Claude Code).

* Launcher: fail closed on a missing download checksum (engine triage C1)

Byte-identical sync of the engine repo's launchers: a freshly downloaded
engine binary now runs only after verifying against its .sha256 sidecar.
A sidecar that cannot be fetched, or a machine with no sha256 tool,
refuses the download instead of exec'ing an unverified binary; the
wget-only path fetches the sidecar too. Binaries already on PATH or in
the cache that pass engine-probe are unaffected.

Prepared with AI assistance (Claude Code).

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WaJv2c4oN8wS7Ttq4XRqyx

* Enforce engine-before-skill release order (triage D4)

The launcher, npm shim, and `impeccable install` all resolve the engine
binary for the pinned ENGINE_VERSION, so a skill/CLI release or a rust-swap
merge published ahead of the engine release + platform packages dead-ends
every install path. Add a mechanical guard:

- scripts/check-engine-release.mjs: verifies all five dist binaries +
  .sha256 and the five @impeccable/cli-<os>-<arch> npm platform packages
  exist for the pinned ENGINE_VERSION; names missing assets, exits non-zero.
  Honors IMPECCABLE_DOWNLOAD_BASE.
- release.mjs: hard-fails release:skill and release:cli when assets are
  missing; extension is exempt (vendored WASM detector, no engine exec).
- CI engine-release-ready job: runs the check, continue-on-error with a
  loud ::warning until the first engine release exists (flip to false then).
- CLAUDE.md Releases: documents the enforced ordering.

Prepared with AI assistance (Claude Code).

* Oracle: re-record the Sep-1 verb fixes ported to the Rust engine

Five fixes landed on main in JS between the swap branch and its rebase and were
ported to the engine; the goldens they touch are re-recorded from the fixed
binary, each engine output first diffed byte-for-byte against the upstream JS on
the same inputs. DELTAS.md documents every case with its upstream hash.

- critique-* (usage/unknown/latest-existing/write-then-read/write-monorepo-child):
  the #660 critique close path (identity + fingerprint freshness, ~NNNN
  collision suffix, closed flag, close verb, latest --json). Upstream 5211bdf4.
- detect-* (new overused-font fixture cases, dir/scope/no-advisory sweeps):
  the #678 overused-font primary-face change (a system stack keeps its system
  face, so a Roboto fallback no longer flags). Upstream 2cfd6076.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WaJv2c4oN8wS7Ttq4XRqyx

* Tests: fix pre-existing release-guard staging on the swap branch

release.test.mjs was already red on the swap branch: release.mjs imports
check-engine-release.mjs and fetch-engine.mjs (the D4 engine release-order
guard), which the temp work tree never staged, so every dry run failed to
resolve the module instead of exercising the guard. Stage both modules and set
IMPECCABLE_SKIP_ENGINE_CHECK=1 so the guard does not probe the network; this
suite predates the guard and only covers the version/changelog/artifact checks.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WaJv2c4oN8wS7Ttq4XRqyx

* oracle: pin E8 stale-hook-manifest detector fallback (context)

Cover the v3-to-launcher upgrade fix (triage E8) recorded from the engine
binary and hand-reviewed:

- context-stale-hook-manifest: a .claude/settings.local.json naming the retired
  `node .../hook.mjs` script under the claude-code provider emits
  MANUAL_DETECTOR_REQUIRED, because the stale marker no longer counts as an
  active hook (its script is gone after the update).
- context-launcher-hook-active: the same manifest in the launcher form still
  suppresses MANUAL_DETECTOR_REQUIRED, confirming the launcher marker is
  recognized as active.

The only difference between the two goldens is the MANUAL_DETECTOR_REQUIRED
block. No existing golden moved: every other context case runs under the source
provider, whose hook-manifest list is empty, so none of them scan a manifest.
Also null IMPECCABLE_PROVIDER_ID in the case BASE_ENV so a recording machine's
value cannot leak. DELTAS.md records the intentional divergence from JS parity.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WaJv2c4oN8wS7Ttq4XRqyx

* Tests: stop two harness hangs from wedging a whole run

Two suites could hang forever and never print a tally, because the one
mechanism that could interrupt the wedged work was missing on both paths.

Hang 1 (bun run test / build-phase.test.mjs): the test's run() helper
spawned every child with spawnSync and no timeout. spawnSync blocks the
test worker's thread, so node's --test-timeout (an event-loop timer)
cannot interrupt a child that wedges (a fork/exec blocked on OS resources
under concurrency, a gate's comp-diff grandchild, or a stray browser
launch). Bound every child with spawnSync timeout + killSignal SIGKILL so
a wedge becomes a fast, named failure the next test survives.

Hang 2 (bun run test:skill-behavior): runTurn called generateText with no
client-side deadline, so a stalled provider stream kept the fetch (and the
whole node process) alive past the per-test timeout, producing no tally.
Attach a real AbortSignal (default 840s, under the 900s per-test cap):
on expiry the fetch aborts, the turn throws, and the scenario
fails-and-continues. The unref'd timer is cleared on completion.

Runner backstops: run-tests.mjs now spawns each command as a detached
process-group leader and enforces a per-suite wall-clock cap that SIGKILLs
the entire group (workers, grandchildren, browsers) on expiry, with
SIGINT/SIGTERM forwarded so Ctrl-C still reaps the tree. The core node
batch gets a finite --test-timeout (180s); skill-behavior gets a 60min
group cap. Env overrides: IMPECCABLE_TEST_WALL_CLOCK_MS,
IMPECCABLE_SKILL_BEHAVIOR_TURN_TIMEOUT_MS,
IMPECCABLE_BUILD_PHASE_RUN_TIMEOUT_MS.

Proof: bun run test green twice (~60s); scoped claude-sonnet-5
skill-behavior sweep terminates with a tally (20 tests, ~32min) where the
840s abort caught a wedged redesign turn and the sweep continued instead
of hanging.

Prepared with AI assistance (Claude Code).

* launcher: export skill-dir env before the IMPECCABLE_BIN exec (sync engine fix)

Prepared with AI assistance (Claude Code).

* Node-free swap: comp-fidelity verbs move to the engine

The four comp-fidelity scripts (comp-spec, comp-diff, font-match, build-phase)
and their six libs are ported into the impeccable-engine binary. This removes
the last Node .mjs from the skill: `git ls-files skill/scripts | grep '\.mjs$'`
now returns nothing.

- reference/new-work.md, reference/visualize.md, and the asset-producer /
  finish-reviewer agents now invoke `{{scripts_path}}/impeccable <verb>` instead
  of `node <script>.mjs`.
- Deleted the ten ported .mjs and the four JS unit tests that imported them
  (their behavior is now covered by the engine's Rust tests and the oracle);
  removed those files from scripts/test-suites.mjs.
- Added oracle cases (comp-*, font-match-*, build-phase-*) over a comp-basic
  workspace, recorded from the engine binary; the deterministic outputs are
  byte-identical to the JS the scripts left behind.
- docs/CLI-CONTRACT.md documents the four verbs, the CDP font rendering, and
  the runtime-resolved (never-committed) font-index catalog.

The font-index catalog JSON stays shipped in the skill (data/font-index.json);
the engine resolves it at run time and never vendors it.

Prepared with AI assistance (Claude Code).

* reorg: public plumbing for the in-repo Rust workspace and the two-release flow

The engine binaries move from the impeccable-dist channel to this repo's own
GitHub Releases (tag engine-v<ENGINE_VERSION>), and the closed detector the
engine links arrives as detector-v<DETECTOR_VERSION> releases on the same
repo. This commit wires the public side for that; the crates themselves land
in the next commit.

- Launcher (sh + cmd), npm shim, fetch-engine and check-engine-release now
  download from github.com/pbakaus/impeccable/releases/download/engine-v<X>/.
- release.mjs gains `engine`: verifies ENGINE_VERSION against the platform
  package pins and the detector release, tags, pushes; release-engine.yml
  builds the five targets and publishes. check-detector-release.mjs is the
  matching release-order guard (with tests).
- Root Cargo.toml (workspace, lto = false with the reason), rust-toolchain.toml
  (exact pin), DETECTOR_VERSION, /target ignored.
- CI: rust + rust-windows jobs and an oracle job that replays the goldens
  against a source build, warn-only until the first detector release exists;
  ci-test-plan exposes a `rust` output.
- docs/ENGINE.md (the crate map and the closed-detector mechanism) and the
  CLAUDE.md engine, release-order and rules sections.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* reorg C: the open Rust runtime joins this repo as one Cargo workspace

The engine no longer lives in a separate repo. `crates/` is a snapshot of the
open crates (foundation, core, common, context, live, hook, skills, comp,
comp-verbs, html, browser, detect, cli) plus `Cargo.lock`, taken as a git
archive of the engine repo at the commit that finished the boundary split.
None of that repo's history comes with it, and none of it should: the closed
half stays private.

The closed half is the rule engine. It ships as a prebuilt native archive per
target, `libimpeccable_detector.a`, published as a `detector-v<X>` GitHub
Release on this repo. `crates/core/build.rs` resolves and links it three ways:
`IMPECCABLE_DETECTOR_LIB=<dir>` for a local detector build, else the
`~/.impeccable/detector/<version>/<target>/` cache, else a download verified
against its `.sha256` sidecar. `crates/core` is a thin shim over a three-symbol
C ABI; nothing above it knows the boundary exists.

What changed versus the engine repo copy:

- Every crate manifest moves from `license-file.workspace` to
  `license.workspace` (this workspace declares Apache-2.0), and the workspace
  gains the `postcard` dependency the boundary encoding needs.
- The launcher contract test reads `skill/scripts/impeccable{,.cmd}` instead of
  a sibling `launcher/` dir, and `engine_binary` downloads from
  `github.com/pbakaus/impeccable/releases/download/engine-v<version>/` instead
  of the retired dist repo. No oracle golden carried the old URL, so no
  re-recording was owed.
- The tests that hunted for a public repo through `IMPECCABLE_PUBLIC_REPO`,
  `../impeccable-second` or a hardcoded home directory now resolve the root as
  `CARGO_MANIFEST_DIR/../..`, because they are in it. The env var stays as an
  override for an out-of-tree checkout.
- The in-page bundle (`detect-antipatterns-browser.js`, 2 MB of generated wasm
  glue) is no longer tracked. `crates/core/build.rs` resolves it beside the
  archive, hands the path to `impeccable_core::browser::IN_PAGE_BUNDLE_JS`, and
  live mode serves that. `scripts/check-detector-release.mjs` now requires it
  and its `.sha256` in a detector release.
- The live crate embeds `skill/scripts/live-browser*.js` and
  `modern-screenshot.umd.js` directly rather than through vendored copies, so
  the binary and the installed skill cannot drift.
- `crates/browser/assets/` (an unused second copy of the bundle) is gone.
- `tests/lib/engine-bin.mjs` also accepts `target/release/impeccable`, so a
  plain `cargo build --release -p impeccable` is enough to run `bun run test`.

Verified with the archive from a local detector build: `cargo test --workspace`
267 pass, oracle 795 pass / 0 fail / 0 missing, `bun run build` clean, the
default suite green, and the launcher's `engine-probe` handshake answering
through `skill/scripts/impeccable`.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* docs: bring RUNTIME-ENV and PORTING-GUIDE over with the runtime

They describe the binary's environment contract and the parity method every
crate here was ported with; both belong next to the crates now.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* core/build.rs: refuse a detector archive built by another rustc, in plain words

The archive links only against the exact rustc that built it; a mismatch
used to surface as pages of undefined std symbols from the linker. The
detector repo now writes rustc-version.txt next to the archive (and ships it
with the release); when it is present, build.rs compares it with its own
compiler and names both versions.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* build:extension: ship the wasm-core extension shell and vendor its detector from the detector release

`bun run build:extension` was broken on this branch: it still imported the
deleted JS engine (cli/engine/registry/antipatterns.mjs,
scripts/lib/browser-detector-bundle.js).

The shipped shell now matches the new design. The content script only
snapshots the DOM; an extension-owned offscreen document runs the
WebAssembly rule core over that snapshot, so the scanned page's CSP no
longer matters. That replaces the old approach of injecting a JS rules
bundle into the page. New files: extension/offscreen/offscreen.html, plus
the "offscreen" permission and a 'wasm-unsafe-eval' extension_pages CSP in
the manifest.

The manifest version stays at 1.3.3. The shell's own manifest carried
2.0.0; feature branches never bump versions, so the bump is a release step.

The five generated detector pieces (core.js, core_bg.wasm, snapshot.js,
overlay.js, antipatterns.json) are vendored at build time into the
gitignored extension/detector/ by the new scripts/lib/detector-bundle.mjs,
which resolves them the same three ways crates/core/build.rs resolves the
native archive: IMPECCABLE_DETECTOR_LIB/extension-detector/, the
~/.impeccable/detector/<DETECTOR_VERSION>/ cache, then a checksum-verified
download of detector-browser-bundle.zip from the detector release.
antipatterns.json is no longer regenerated here.

The zip packaging is unchanged. The Firefox variant still builds so
`web-ext lint` keeps covering the shared shell, but it cannot scan: Gecko
has no chrome.offscreen API. The build prints a one-line warning saying so.

Also here: a referenced-path check that fails the build when the manifest
or the service worker points at a file that is not in extension/, a
resolver unit test wired into the core suite, and the detector rule count
in the READMEs synced to the 61 the vendored registry carries.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* oracle: replay byte-for-byte on Linux too

The corpus was recorded on macOS and eight cases failed on ubuntu CI for
reasons that were all environment, not behavior:

- stageWorkspace returns the realpath of the staged dir. macOS's tmpdir is a
  symlink and two goldens (context-dir-override, live-accept-source-locked)
  had recorded that artifact; both re-recorded, reviewed in DELTAS.md. The
  source-locked case now actually exercises the lock it is named for.
- context-lowercase-product-name declares platforms: ['darwin', 'win32'];
  run.mjs skips such cases elsewhere and says so in the summary.
- The hook-project workspace's empty provider skill folders (.claude,
  .cursor) are now tracked with .gitkeep; git cannot track empty
  directories, so a fresh checkout had none and hooks on found nothing to
  repair.
- crates/live's read_dir_raw sorts entries by name: the goldens hold the
  order macOS returned, Linux returns hash order, and the source-candidate
  lists in live-commit output depended on it.

macOS: 795 pass, 0 fail. The Svelte accept cases additionally need the
public repo's node_modules on the machine that runs them (CI now installs
them).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* oracle: mask <HOME> only at path boundaries (a short home like /root ate 'roots.json')

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* oracle: track live-html's dist/generated.html (the root dist/ ignore hid it from CI checkouts)

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* release-engine: darwin-x64 builds on macos-14 (macos-13 is retired)

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Open the detector: the rule crates join the workspace, the C-ABI goes away

The detector is open source. The rules it ships were already public in this
repo's git history and in every npm tarball of the JS engine, so a closed
binary bought nothing it could keep; the moat is the service (the catalog,
the labs, the review pipeline), not the check functions. Keeping them behind
a prebuilt archive cost a C-ABI, an exact toolchain pin, a build-time
download, a second release to order ahead of every engine release, and a
serde layer that had to serve two encodings.

Deleted
- crates/core/src/ffi.rs, crates/core/build.rs, crates/core/tests/boundary.rs
  and the shim modules under src/checks and src/browser.
- crates/foundation/src/boundary.rs and the postcard dependency.
- DETECTOR_VERSION, scripts/check-detector-release.mjs and its test, the
  check:detector-release script, the detector gate and
  IMPECCABLE_SKIP_DETECTOR_CHECK in scripts/release.mjs.
- scripts/lib/detector-bundle.mjs and tests/detector-bundle.test.mjs (the
  vendoring path for the closed browser bundle).
- scripts/build-browser-detector.js and the build:browser script (a stub
  since the JS engine left the tree).
- xtask's detector-archive subcommand and its public-repo lookup.

Came back
- crates/core is now the rule logic itself: every check_* / scan_*, the
  browser adapters, the visual-contrast decisions. It re-exports foundation
  as before, so no consumer changed. Its vectors dispatcher is the union of
  both id tables again, and tests/vectors.rs replays the frozen vectors
  straight through it.
- crates/wasm and crates/xtask join the workspace. cargo xtask bundle builds
  the in-page bundle from browser-bundle/ plus the wasm core, writes
  dist/, refreshes the tracked crates/live/assets/detect-antipatterns-
  browser.js, and writes extension/detector/. bun run build:extension runs
  it instead of downloading.
- crates/live/assets/detect-antipatterns-browser.js is tracked again; live
  mode embeds it and serves it as /detect.js.
- Serde is back to plain derives: no is_human_readable branch in
  js::json_number, derived Serialize for Rgba and BrowserFinding with their
  skip_serializing_if attributes.
- profile.release has lto = "fat" again; rust-toolchain.toml is plain
  stable plus the wasm32 target. The rust, rust-windows and oracle CI jobs
  lose continue-on-error and can be required.

Verified
- cargo build --workspace --all-targets: clean, no warnings.
- cargo test --workspace: 346 pass, 0 fail (the 8 boundary tests are gone
  with the boundary).
- cargo build -p impeccable-wasm --target wasm32-unknown-unknown --release: ok.
- cargo xtask bundle && cargo xtask bundle --check: reproducible; the
  regenerated bundle is committed (it differs from the archived one, which
  was built with a pinned rustc and lto = false).
- cargo build --release -p impeccable: no linker warnings, 12.5 MB (the
  same source at lto = false is 13.1 MB).
- oracle: 795 pass, 0 fail, 0 accepted deltas, 0 missing goldens.
- bun run build, bun run build:extension, web-ext lint (0 errors,
  8 warnings), bun run test: 363 + 80 + 1 + 1 + 133 + 180 + 4 pass, 0 fail.
- impeccable detect --no-config --json tests/fixtures/antipatterns: 128.7 ms
  median of 5.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* core: doc comments drop the open/closed split

The rule crate and the foundation crate are both Apache-2.0 in one
workspace now, so "open", "closed" and "crosses the boundary" no longer
describe anything. Comments only.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Rule packs: downstream crates add rules on all three engines; wasm detect surface

A crate that depends on this workspace can now add rules without forking
it. `impeccable_core::rule_pack::RulePack` (object-safe, Send + Sync +
Debug) carries a pack's registry rows plus three hooks that default to
empty: `check_text` for the text engine, `check_element_dom` and
`check_page_dom` for the browser driver. `impeccable_html::StaticRulePack`
adds `check_document` for the static engine, where the document model
belongs to the html crate and detect cannot name it.

The registry keeps ANTIPATTERNS as the built-in list; `registry::extend`
appends a pack's rows and every lookup consults them after the built-ins,
so a pack can never shadow a built-in id (extend panics on a collision and
is idempotent per slice). `all_antipatterns()` is the built-ins followed by
the registered rows.

Hook order, chosen so built-in output cannot move:

- detect_text: after every matcher, analyzer and the dedupe, before inline
  ignores, so `impeccable-disable` waives pack rules like built-in ones.
- detect_html_source: after the element rules, the design-system merge and
  the page passes, again before inline ignores. One pack pass per HTML
  file: the document hook when set, otherwise the text hook over the raw
  source, so a pack implementing both never reports twice.
- collect_browser_findings: the element hook at the end of the per-element
  loop through the same disabled-rules filter and group, the page hook
  after every built-in page pass with the same el-or-body attribution.

A pack travels on TextOptions / ScanOptions, DetectHtmlOptions
(static_rule_pack plus rule_pack), StaticHtmlEngine, and BrowserConfig
(serde-skipped: a pack is a Rust value, not JSON from the page). The
shipped binary installs none.

`crates/wasm --features detect` exposes the two file engines as JSON
exports for hosts that cannot exec the binary: `detect_text_json` and
`detect_html_source_json`, options `{ inlineIgnores?, designSystem? }`,
returning the findings array `detect --json` prints. `antipatterns_json`
now includes a pack's rows. `set_rule_pack` and `set_static_rule_pack` are
Rust-only, for a crate that links this one as an rlib.

Tests: registry extension and collision in foundation, one test pack per
engine (crates/core, crates/detect, crates/html tests) proving each hook
fires, that the built-in findings are unchanged, and that the waivers and
the disabled-rules list cover pack rules, plus the wasm export shapes.
Workspace tests 346 to 361, oracle 795/0 unchanged.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* docs: the cutover checklist under the open design

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* bundle: the page JS and the bundler become a library crate downstream packs can reuse

The in-page bundle, the extension pieces, the registry JSON and the
wasm-pack call were reachable only through `cargo xtask bundle`, which read
`browser-bundle/*.js` from the repo root. A downstream crate that links
impeccable-core + impeccable-wasm with its own rule pack had to copy the
page JS to produce a detector bundle for its module.

They move to `impeccable-bundle` (crates/bundle), which embeds every
`browser-bundle/*.js` with `include_str!` and exposes `in_page_bundle`,
`extension_pieces`, `registry_json`, `check_capture_contract` and
`wasm_pack_build`. Nothing writes files or exits the process; the caller
places the bytes. `registry_json` now reads `all_antipatterns()`, so an
installed pack's rows land in `antipatterns.json` too (no built-in change).

xtask becomes the workspace's caller and writes the same files to the same
places; `cargo xtask bundle` is byte-identical, tracked live asset included.
`IMPECCABLE_BUNDLE_SKIP_WASM_PACK` is the skip switch's new name, the old
`IMPECCABLE_XTASK_SKIP_WASM_PACK` still works.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* The immediate tier moves to the registry, and reaches wasm

The design hook's immediate-tier list is the set of rule ids worth fixing
at the edit site, and a downstream reviewer wants the same set to decide
how loudly a finding is reported. `impeccable-hook` is native-only, so the
list moves to `impeccable_core::registry` (the hook re-exports it) and the
`detect` feature gains `immediate_tier_rules_json()`.

The export is behind `detect`, which the in-page bundle does not build, so
the tracked browser asset is unchanged.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* docs: Pristine tracks the engine by revision pin, not npm

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* docs: the cutover checklist is maintainer-side, not part of the tree

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: Fix flat type hierarchy false positives (#702)

Upstream sha 84728e9ce4.

The rule now reads rendered semantic roles and the dominant size per role
instead of the raw set of font sizes on the page, and it fires only when
every adjacent role step is under 1.25x.

- crates/core checks::rules gains TYPE_HIERARCHY_SELECTOR / MIN_ROLES /
  MIN_STEP_RATIO, typeHierarchyRole, dominantTypeRoleSize and
  checkFlatTypeHierarchySamples, the shared half of checks.mjs.
- crates/core browser::page_checks gets checkFlatTypeHierarchyFromDoc over
  the Dom trait, with the overlay skip selector checkTypography passes.
- crates/html page.rs gets the same walk over StaticDocument.
- crates/detect drops the source-only analyzer: flat-type-hierarchy leaves
  REGEX_ANALYZERS, the text-content analyzers shift to index 1, and
  analyzer_rule_id loses its first row.
- crates/html cascade defaults gain contentVisibility, and crates/foundation
  registry carries the reworded description.

Goldens re-recorded (the binary now matches origin/main's JS engine on every
one of these fixtures, verified by scanning the shared corpus with both):
glow, icon-tile-stack, layout, modern-color-borders, motion,
named-color-borders, numbered-section-markers, oklch-neon-text,
typography-should-flag, json and text.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: Fix detector URL scans and advisory handling (#709)

Upstream sha fa44839f72.

Advisory handling. `severity` becomes the canonical registry field: the
`advisory` bool leaves `Antipattern`, `advisory_rule_ids` filters on
`severity == "advisory"`, and `derive_advisory_flag` stamps the finding's
`advisory: true` from the effective severity, so a per-finding promotion or
demotion carries the flag. The html and browser engines call it after their
severity override; the detect CLI and the hook accept either spelling; the
driver's serializer and the wasm registry exports derive it the same way.
em-dash-overuse moves from `advisory: true` to `severity: "advisory"`.

URL scans. `expand_joined_url_targets` splits an argv value that is entirely
whitespace-separated URLs and leaves paths with spaces alone. The browser
driver reads the readable linked-stylesheet corpus into the HTML pattern
corpora and resolves a finding's selector with `selector_nodes_for_live_dom`
/ `pseudo_element_host_selector`, so an unresolvable selector drops the
finding instead of keeping it page-level. The CSSOM walk itself is page JS:
`browser-bundle/15-snapshot.js` gains `__snapLinkedStylesheetText` (grouping
rules flattened, container-query probes, effective keyframes) and puts it in
the snapshot as `linkedCss`; `10-probe.js` exposes the same for the in-page
route, and the Dom trait carries `linked_stylesheet_text`.

Also `enclosing_css_selector` blanks comments before hunting the previous
declaration delimiter, and `check_typography` reports the uniquely most-used
family instead of every family over a 15% share.

Verified: `impeccable detect --no-config --json tests/fixtures/antipatterns`
is now byte-identical to `node cli/bin/cli.js` on an origin/main worktree
over the shared corpus (432 findings). The two changed lines in
tests/oracle/vectors/calls/rules.checks/checkHtmlPatterns.jsonl were
re-recorded by running origin/main's `checkHtmlPatterns` over the frozen
args; only the comment-polluted selector changed. Goldens re-recorded for
the advisory partition (config-*, fixture gemini/gpt-tells,
numbered-section-labels, scoped-ignore, shape-assembled-illustration,
color, em-dash-entities) and the help text, each cross-checked against the
JS on origin/main.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: stop gray-on-color false positives on Tailwind opacity and JSX (#707)

Upstream sha 32b270f4e8.

`find_solid_chromatic_bg` replaces the bare `bg-<hue>-<n>` match in both
engines: a `bg-blue-500/10` tint is a wash, not a solid fill. The `regex`
crate has no lookahead, so the maximal digit run plus the word boundary is
matched as before and the byte after it is tested for `/`.

The text engine gains the JS-source scanner (`scan_js`) and the scope
helpers on top of it: `containing_markup_tag` keeps a gray text class from
pairing with a background in a sibling tag on the same line, and
`find_ternary_split` / `exclusive_class_scopes` split a `cond ? a : b`
class expression into its arms, recursing into nested ternaries, ignoring
`?.` and `??`, and keeping a common prefix and post-ternary suffix in every
arm. `MatchCtx` now carries the match offset the scope lookup needs.

Verified against origin/main's JS: all eleven cases from the upstream test
file plus a nested / nullish / suffix set produce byte-identical findings on
both engines; they are pinned as Rust unit tests in `regex_matchers` and
`checks::rules`. The shared fixture corpus stays byte-identical.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: resolve unique --target names in monorepos (#706)

Upstream sha 8b326fc81e.

`resolve_target_path` / `find_unique_bare_target` in `crates/context`: a
`--target` that does not exist and reduces to a single path segment under
cwd resolves to the one workspace candidate with that name, so `--target a`
selects `apps/a`. A caller that already absolutized the name against cwd
(live and the other helpers do) takes the same route. Ambiguous or unknown
names still report the miss.

The context CLI resolves the target once and hands the resolved path to
`load_context`, replacing `path_exists_for_target`.

Oracle: four new `context-monorepo-target-bare-*` cases (bare name,
absolutized bare name, unknown name, bare name from a child cwd).
`context-monorepo-target-b-inherits` was re-recorded: resolving the target
before `load_context` changes its `surfaceBriefReason` from `not-found` to
`invalid-target`, which is what origin/main's `context.mjs` prints for the
same run.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: Fix Next.js 16 CSP and parent hook discovery (#710)

Upstream sha 672ca29642.

CSP. `detect-csp` recognizes Next.js 16's `proxy.{ts,js,mjs}` request hook
beside `middleware.*`, but only where it sits at a project root or its `src/`
directory: the scan root itself, or a nested directory carrying a Next
project marker (a `next.config.*`, an `app` / `pages` dir, or a `next`
dependency). A same-named helper elsewhere in the tree is not the framework
hook.

Context. `find_git_boundary_root` gives `resolve_project` a git-boundary
notion: an explicit target inside its own repository resolves against that
repository, and an external target resolves against its own root, so caller
context never leaks across the boundary. `hook_manifest_search_roots`
replaces the cwd/projectRoot/repoRoot triple with a walk up from the
project root that stops at the first git boundary, and each root's own hook
lifecycle config is honored before its manifest counts as coverage.

Verified against origin/main's JS: nine `detect-csp` placements and five
hook-discovery scenarios (enclosing harness root, that root disabled,
sibling target, nested git target, markerless nested git target) produce
identical output.

Oracle: five `csp-proxy-*` cases and five `context-hook-*` /
`context-markerless-nested-git-target` cases. Four route-target goldens were
re-recorded because #710 resolves a `/`-prefixed target outside the
workspace; each was cross-checked against origin/main, and
`surface-brief-write-route` has a DELTAS entry for the one wording
difference (an unwritable filesystem root).

`tests/framework-fixtures.test.mjs`'s new proxy-placement block came in from
the merge importing the deleted `detectCsp`; it now drives `detect-csp`
through the binary like the rest of that file.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: fail URL scans when the browser is unavailable (#711)

Upstream sha f2f9958be1e6a4ecb1fbd5ef1ae1b7d9c53e0d24 (Fix: fail URL scans
when the browser is unavailable).

`detect` gains an operational-failure flag. Exit 1 now means at least one
requested target could not be scanned, and it takes precedence over exit 2,
because findings from the targets that did scan do not turn a partial scan
into a complete one. The flag is set by an unreachable path, an unreadable
directory or file in a dir walk, a per-file scan that throws, a URL scan
that throws, and a shared-browser setup failure.

- `walk_dir_reporting` and `build_import_graph_reporting` take a read-error
  callback; the plain wrappers stay for callers that do not report. A file
  the graph could not read is skipped for the scan too.
- `SharedBrowser::ensure_launched` is the eager half of
  `createBrowserDetector()`: the CLI brings the browser up before the loop so
  a launch failure prints one `Error:` line and every URL target is skipped,
  instead of the lazy launch reporting once per URL.
- The static engine and the text path spell a permission failure the way Node
  does (`EACCES: permission denied, open '<path>'`), which is what
  `Error: cannot scan <target>: <message>` prints.
- Usage text and docs/CLI-CONTRACT.md carry the exit-status block.

Verified against origin/main's JS: missing target, missing target alongside a
flagging file, unreadable file, unreadable file beside a readable sibling,
unreadable directory, unreadable nested directory, a clean scan, and a
browser-unavailable scan of one and of two URLs all agree on exit code,
stdout and stderr (the browser-not-found wording is the pre-existing
puppeteer-vs-discovery difference).

Oracle: `detect-missing-file` and `detect-missing-file-json` re-recorded at
exit 1, plus new `detect-missing-file-with-findings`,
`detect-unreadable-file-json` and `detect-unreadable-file-in-dir`, each
cross-checked against origin/main. `detect-help` carries the new block.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: OpenCode slash command bridge (#483)

Upstream sha 9736a9f6e9.

OpenCode does not honor `user-invocable: true` on SKILL.md frontmatter, so a
pinned skill never reaches its slash menu. `pin` now writes
`commands/impeccable-<cmd>.md` on the OpenCode command schema instead, and
skips `.opencode` in the SKILL.md loop so no unreachable
`.opencode/skills/<cmd>` is left behind. `unpin` mirrors it, marker-guarded,
and reaches both scopes even when the skill itself is gone.

`find_opencode_commands_dirs` covers the project-local dir when the project
has the skill and the user config dir when Impeccable is installed globally,
resolving that dir the way the CLI does (`OPENCODE_CONFIG_DIR` ->
`XDG_CONFIG_HOME/opencode` -> `~/.config/opencode`).

The build-tooling half of the upstream change (transformers, the OpenCode
command the build generates, `root-commands-sync`) came in with the merge and
needed no port.

Verified against origin/main's pin.mjs across seven scenarios (no harness,
project scope, user scope, a foreign command file, pin then unpin, unpin over
a foreign file, unpin with nothing pinned): identical stdout, identical file
sets, identical file contents apart from the one deliberate difference.

Oracle: five `pin-opencode-*` cases, with a DELTAS entry for the bridge body
naming the launcher rather than `node .../context.mjs`.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: Fix Codex skill version metadata (#703)

Upstream sha 482368511a.

Codex's validator rejects unknown top-level keys, so the Codex and `.agents`
skills now carry `version` under the spec-defined `metadata:` map. Both
version readers learn the same parser: `parse_skill_frontmatter_version` in
`crates/context` (the boot update check) and `extract_version` in
`crates/skills` (`getSkillsVersion`). A metadata version wins, a legacy
top-level one still reads, only the map's own indent level counts, tabs count
as two spaces, and a comment line is skipped.

The build-tooling half (`versionInMetadata` on the two providers, the YAML
emitter's nested-object branch) came in with the merge.

Fourteen frontmatter shapes were recorded from origin/main's
`parseSkillFrontmatterVersion` and pinned as unit tests in both crates.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: Fix skill subcommand help handling (#708)

Upstream sha a264199177.

`install`, `link`, `update` and `check` render static help before entering
any operational path, through both the top-level verb and the legacy `skills`
namespace, for `--help` and `-h` alike.

Verified against origin/main's `cli/bin/cli.js`: all six spellings produce
identical text and exit codes.

Oracle: a new `tests/oracle/cases/skills.mjs` with seven help cases. Only the
help paths are pinned there; every other installer path writes into harness
directories or reaches the network.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Oracle: goldens for the three fixtures the merge added

`tests/fixtures/antipatterns/` gained `flat-type-hierarchy.html` (#702) and
`linked-url-patterns.{css,html}` (#709) with the merge, so the corpus
generator produced six `detect-fixture-*` cases with no goldens and the
directory-wide cases (`detect-dir-*`, `detect-scope-*`, `detect-no-advisory-*`)
moved.

Every golden here was recorded from the binary and then cross-checked against
`node cli/bin/cli.js` on an origin/main worktree over the same files: the six
per-fixture cases agree byte for byte in JSON and text, and a full scan of
`tests/fixtures/antipatterns` produces 432 findings identical on both engines
after normalizing the repo path.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Port: the installer half of the OpenCode command bridge (#483)

Upstream sha 9736a9f6e9, the part of it that
lives in `cli/bin/commands/skills.mjs` rather than `pin.mjs`.

`copy_provider_commands` mirrors `copy_provider_skills` for a provider's
compiled `commands/` dir: project scope writes `<root>/<configDir>/commands`,
user scope writes the config dir OpenCode actually scans
(`OPENCODE_CONFIG_DIR` -> `XDG_CONFIG_HOME/opencode` -> `~/.config/opencode`),
and a pre-#406 global install at `~/.opencode/commands/` loses exactly the
files just written while siblings, symlinked dirs and home-rooted git repos
are left alone. It runs on install, on the reinstall refresh, on update, and
on link, which is the only path that can deliver the bridge to a linked
install.

`is_up_to_date` now compares the bundle's command files too, so an install
whose skills match but whose bridge is missing or drifted refreshes instead of
reporting success while the slash command stays absent. Only bundle-shipped
files are compared, so a pinned shortcut never affects freshness.

`tests/copy-provider-commands.test.js` arrived with the merge importing the
deleted `cli/bin/commands/skills.mjs`; its scenarios are ported to
`crates/skills/tests/provider_commands_tests.rs` (project scope, the three
user-scope dir resolutions, the legacy migration and its two guards, a
provider with no commands dir, and the four `isUpToDate` command-awareness
cases), and the file is removed and deregistered.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* CI: the first full run on the branch, three fixes

- The oracle harness masks the climb to the root a /-prefixed target
  produces (<UP_TO_ROOT>/): the number of `../` is the staged tmpdir's depth
  (7 on macOS, 2 on Linux), not the verb's behavior. surface-brief-path-slash
  re-recorded.
- Two context test helpers canonicalized their temp dir, which on Windows
  yields a \\?\ verbatim path that takes `/` literally; they strip the prefix
  like Node's realpathSync. The critique-storage identity test compares
  against the platform's own resolved path.
- Every job that drives the binary end to end (live-e2e smoke and full,
  accept-cleanup, the DeepSeek sweep, the remote CLI smoke) builds it from
  the checkout first; before, they looked for a release that does not exist.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* context tests: the verbatim-prefix strip spells the prefix once

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* context test: derive the snapshot identity from the verb's own resolver

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* context test: JSON-quote the snapshot identity, as the verb does

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* detect test: import resolution against platform-form paths

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* hook test: the stock cache path in the host's path form; Windows CI runs every crate's tests before failing

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: skills tests pass on Windows

The two test temp roots kept `canonicalize`'s `\\?\` verbatim prefix, and the
kernel takes a verbatim path literally, so every `/`-joined path built under
them was an invalid filename. Strip it the way Node's `realpathSync` does.
The manifest, artifact and sibling-binary expectations hard-coded POSIX
separators for paths the product joins with the host's semantics; derive them
from `jsp::join` instead.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: hook tests pass on Windows

Same verbatim-prefix strip on the test temp roots, plus expectations derived
from the helpers the product uses: cache keys and scan targets from
`jsp::join`, the config path in an admin message from the same relative form
`path.relative` renders, and the footer hints from `quote_command_arg`, which
deliberately switches to the double-quoted Windows form (#476 / #533). The
env lock no longer poisons the sibling tests when one of them fails.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: html oracle goldens compare on Windows

The goldens pin the `<REPO>`-masked fixture path recorded on POSIX. Mask, then
render the remainder with `/` so a Windows checkout's backslashes are not read
as a finding difference. The goldens are untouched.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: widen the live read-deadline test's margin

Timing only. The watchdog polls in 50ms steps against a ~15.6ms Windows system
timer while the crate's tests run in parallel, so the later request takes its
turn later there. The bound stays far under the 60s read timeout a
deadline-less read would hold the ticket for, so the test still distinguishes
the fix from the regression.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: the request read deadline was not enforced on Windows

Windows does not unblock a `recv` already parked in the kernel when another
thread calls `shutdown` on the same socket, so the watchdog could not end a
silent connection's read and it held its turnstile place for the whole 60s
header timeout instead of the 10s deadline. Bound the read at the socket too,
which enforces the same deadline everywhere; the watchdog stays as the backstop
for a connection that trickles bytes without ever completing a request. POSIX
behavior is unchanged: the watchdog already closed the socket at the deadline.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: hook tests derive the rest of the host path forms

The test temp helper's `write` returned a `PathBuf::join` result, which keeps
the `/` inside the relative part and so does not match what the hook resolves a
relative target to on Windows. Three more admin messages and the cache-root slug
pinned the POSIX spelling of paths the product renders with the host's
semantics (`path.resolve` also prefixes the current drive there).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: skills test fixtures name USERPROFILE, and the win32 quoted form

`os.homedir()` reads USERPROFILE on Windows, so a fixture home that named only
HOME sent the global installs into the runner's real profile. The Windows hook
command carries the JSON-quoted path, so a host path's backslashes arrive
escaped; derive the expectation instead of pinning the POSIX spelling.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: check the oracle fixtures out with LF

A finding's snippet carries the scanned file's own bytes, and the goldens were
recorded from a POSIX checkout, so a CRLF checkout of a linked stylesheet reads
as a finding difference. The goldens are untouched.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* windows: check the grok global-install manifests as JSON

The Windows hook command carries the JSON-quoted launcher path, so the path's
backslashes are escaped once inside the command and again by the manifest file
itself. Read the manifest as JSON and look for either quoting form instead of
counting escaping layers in a raw substring match.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* npm shim: refuse a download with no verifiable sidecar

The skill launcher and `impeccable install` both fail closed when a
release binary's `.sha256` sidecar cannot be fetched or carries no hash:
they refuse rather than cache an unverified binary. The npm shim did not.
It only compared when a hash was present, so a 404, an empty sidecar, or
a truncated one all wrote the payload straight into
`~/.impeccable/bin/<version>/` and exec'd it.

It now refuses in the same cases, with wording that matches the launcher,
and writes nothing until the hash matches, so a refusal leaves the cache
dir empty. IMPECCABLE_BIN and the optional-dependency lookup are
untouched: neither downloads.

tests/cli-shim.test.mjs runs the real shim against a throwaway HTTP
server and covers missing, empty, and mismatched sidecars, plus the
matching-sidecar and IMPECCABLE_BIN paths. The two refusal cases fail
against the old shim.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Oracle fixture: declare the vite plugin the web workspace imports

`live-workspaces/apps/web/vite.config.js` imports `@vitejs/plugin-react`
but the workspace's package.json listed only `vite`. No oracle case
installs or evaluates that config (the three `live-boot-workspaces-*`
cases stop at root resolution), so the fixture was never wrong at
runtime, only self-contradictory to read. Adding the devDependency keeps
the goldens byte-equal.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Vectors: drop the 12,208 byte-identical repeat lines

The recorder deduplicated by arguments per run, not across runs, so the
frozen call snapshot arrived with 12,208 lines (43% of 28,266) that
repeat an earlier line byte for byte. Every one re-asserts what its first
occurrence already asserts, and `crates/core/tests/vectors.rs` replays
line by line with no count anywhere, so removing them changes nothing it
checks: the replay still reports 8,321 pass, 0 fail.

Duplicates were removed with `awk '!seen[$0]++'`, keeping first
occurrences and file order, and every changed file was checked to equal
that transform of its old contents. No line was added, reordered, or
rewritten, and no vector file gained or lost a distinct call. The tree
drops from 9.2 MB to 5.7 MB.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Fix: restore the live overlay's disabledValues waivers in the engine

The JS engine applied value-level ignore waivers at the tail of
collectBrowserFindings: `_disabledValues` read the entries the live
overlay resolved for the page (skill/scripts/live-browser-ignores.js
sends them as config.disabledValues), and filtered the assembled
findings by the value each one reported, with design-system-color
compared by color value rather than by spelling so a hex waiver
suppressed a finding the browser reported as rgb(...). The Rust port
dropped that stage: `disabledValues` appeared nowhere in the workspace
or in browser-bundle, so a project entry like

    [detector]
    ignoreValues = [{ rule = "overused-font", value = "geist mono" }]

stopped reaching the overlay. The rules the CLI and the edit hook waive
kept drawing markers and counting toward the badge.

Restore it end to end:

* BrowserConfig gains `disabled_values`, parsed leniently so a
  hand-edited __IMPECCABLE_CONFIG__ entry of the wrong shape is dropped
  rather than failing the whole config, the way the JS filter did.
* The driver applies the waivers after every pass, so a rule pack's
  findings are covered the same way the built-in ones are, honoring the
  entries only in extension mode exactly as the JS read them. The
  normalizer, the value extractor (including the rule that bounce-easing
  without a direct ignoreValue offers no value) and the hex/rgb color
  key are ported alongside it.
* collectConfigJson in the in-page bundle and configJson in the
  offscreen bundle forward the field. The extension never sends it, so
  its behavior is unchanged.

Coverage: two driver unit tests (suppression by font value, by hex
waiver across the rgb spelling, and the extension-mode gate; plus the
config parse and the normalizers), a skipScan test that pins the empty
shape for every stage the core produces, and
crates/wasm/tools/disabled-values-check.mjs, a browser-backed check
ported from the retired tests/detect-antipatterns-browser.test.mjs case
that the swap left without a replacement. Against the previous bundle it
fails on exactly the three waiver assertions and passes the skipScan
one, which is the shape of the regression.

Two related review findings were checked and are not defects. skipScan
is gated on extension mode in both the driver and the bundle, which is
what the JS did (index.mjs#skipScanActive), and the live overlay runs in
extension mode: live-browser.js sets `s.dataset.impeccableExtension` on
the injected /detect.js tag, and the overlay's whole detect toggle
travels over the postMessage loop that 50-scan.js installs only under
EXTENSION_MODE. The visual contrast stage is not leaking either:
collectBrowserFindingsAsync and scan() both consult skipScanActive(),
and the offscreen path skips its visual pass on config.skipScan.

The tracked live asset is regenerated (cargo xtask bundle). The oracle
replays with zero unreviewed differences: the new field defaults empty
and the filter is inert without it, and no CLI path sets extension mode.

AI-assisted change: implemented with Claude Code under maintainer
direction.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Shim test: run from a staged copy and prove the download happened

The three fail-closed cases cleared IMPECCABLE_BIN and pointed
IMPECCABLE_HOME at a temp dir, but locate() prefers an installed
@impeccable/cli-<os>-<arch> before the cache or a download. Those
platform packages ship with every engine release and are a merge
prerequisite, so as soon as one is installed under the repo the cases
would resolve it and go green without fetching anything. Confirmed by
hand: with a platform package staged in node_modules, running the shim
against an unreachable download base still exits 0 from the package.

The shim now runs from a throwaway copy at <tmp>/cli/bin/cli.js beside a
copy of the repo's package.json, with no node_modules on the lookup path
above it, so require.resolve of the platform package fails the way it
does on a machine without the optional dependency. Production code is
unchanged; there is no test-only branch in the shim.

The fixture server also records every request now, and each download case
asserts the asset and sidecar URLs were actually requested, so a future
lookup shortcut fails loudly instead of passing on an untested path. A
sixth case installs a fake platform package next to the staged shim and
asserts the shim prefers it with the server untouched, which pins the
precedence the other cases depend on being absent.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Live: the loader now hands off when the resume is the arrival

The overlay could sit in its generating shader over a DOM that already
held all three variants, and only a page refresh cleared it (#719).

The server's generation preflight runs live-wrap with
--defer-source-write, so the wrapper and every variant reach the DOM in a
single HMR batch. The deferred-wrapper scout is constructed at init and
the variant MutationObserver at Go; observer callbacks run in
construction order, so on that batch the scout resumes first and
resumeSession, not the observer, is the transition into CYCLING. It set
the state and the bar but never called hideShaderOverlay(), so the frozen
capture of the original stayed painted over the variants. It also
reported browser_resumed, which does not count as publication progress,
and then disconnected and re-created the observer, dropping the records
that observer had already queued for the same batch, so variants_ready
never fired at all.

resumeSession now finishes the same transition the observer does (shader
down, inline edit off, insert session finalized, params panel rebuilt)
and reports variants_ready when it already holds every variant. The
deferred scout names itself in the journal as
browser_resumed_deferred_wrapper, so the two resume paths are no longer
indistinguishable.

Wrapper resolution goes through findVariantsWrapper, which prefers a
wrapper that actually holds non-original variants. A target inside a
.map() renders one wrapper per item, and an agent that relocates the
wrapper out of the shared primitive live-wrap scaffolded leaves an empty
one behind; first match could pin either and strand the session at 0/N.
With zero or one match this is the querySelector it replaces.

Tests: waitForCycling now asserts the generating shader is gone once the
bar cycles, across every runtime fixture (it failed on vite8-react-plain
before this change and passes after), marked no-retry so the reload
recovery cannot hide it. Source-shape tests pin the transition, the
variants_ready report, and the wrapper preference.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Live server: stop ends the process, SSE skips the mutation lane

Two Rust-only regressions found while investigating #719, both of which
can leave a tab waiting on a broadcast that never comes.

/stop ran shutdown() but never set shutting_down, and the accept loop
only breaks on that flag or a signal, so a stopped server kept its port
and kept answering while its server.json was already deleted. The next
`impeccable live` then booted a second server on another port and a tab
could reattach to the zombie. Node's shutdown() ended in process.exit(0).
The flag is now set after the response is written, so `stop` still reads
"stopping" instead of a reset connection, and the accept loop (already
non-blocking) exits on its next pass.

GET /events took a turnstile ticket and waited its turn before
registering, even though handle_sse releases that ticket two statements
later and needs no arrival ordering. A peer that stalls mid-request holds
the lane for the whole READ_REQUEST_DEADLINE, so a reconnecting stream
could sit unregistered for up to 10 seconds (measured 9.71s against 0.00s
on Node); broadcast is fire-and-forget, so a `done` landing in that
window reaches an empty client set and is gone. Registering early can
only make a stream see more broadcasts. The one cost is that the
connected frame's activeSessions snapshot may miss a mutation still in
flight, and the browser treats that snapshot as a hint. Preflights still
take a turn: answering those out of order reorders the POSTs the browser
issues behind them.

The route classification moved into releases_ticket_up_front so it can be
unit tested. tests/live-server-leak.test.mjs gains a guard that a stopped
server's pid is gone and its port is free.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Live: the shader teardown can no longer race its own construction

The new cycling assertion caught a real defect on CI: vite8-react-insert
reached CYCLING with #impeccable-live-shader still painted over the page.

showShaderOverlay is async. It appends its canvas synchronously, then
awaits createImageBitmap and finishes the GL setup before it publishes
shaderState. hideShaderOverlay returned early on a null shaderState, so a
teardown that landed inside that window did nothing, and the construction
then published itself over a session that had already left GENERATING,
with no teardown left to run. The scroll tick kept repositioning it,
which is why the CI page.html shows the canvas sized from the capture
rect but styled to the cycling anchor.

Every teardown now bumps a shader epoch before it does anything else, and
a construction pins the epoch it owns and abandons its canvas (releasing
the GL context) at every point past an await and before any publish,
including both bitmap-fallback publishes. A teardown also drops a shader
node that no shaderState owns, so an already-orphaned canvas cannot
survive one.

Reproduced by widening the append-to-publish window: with a 400ms delay
after uiAppend, vite8-react-insert failed with the CI error and the probe
showed the teardown arriving at CYCLING with shaderState still null.
The same run passes with this change, as does a 1500ms window on insert
and plain. Locally that window is about 4ms, which is why it only showed
on a slower runner.

The four remaining setLiveState('CYCLING') sites that did not lower the
loader now do: the SSE done handler (the one route that can reach CYCLING
from GENERATING), the Svelte republish remount, and the two accept
failure recoveries.

The e2e assertion already waits up to 5s for the shader to clear, so it
was never racing a legitimate teardown; it is left as it is.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Live: every active-session wrapper lookup goes through the resolver

Cursor Bugbot on #720: findVariantsWrapper alone was not enough.
resolveBarAnchor, the visible-variant element, mountedParameterCount,
readVisibleVariantFromDOM, showVariantInDOM, the source injection, and
the whole accept path still took the first [data-impeccable-variants]
match, so in the relocated-wrapper case Tune never bound and the bar kept
anchoring to the empty scaffold even after the resume reached CYCLING.

Thirteen call sites now resolve through findVariantsWrapper. The resolver
split in two so a missing id cannot silently widen the lookup to any
session: findVariantsWrapper(sessionId) returns null without an id, and
findAnyVariantsWrapper() is the entry point for the two resume paths that
have no id yet. Both share pickPopulatedVariantsWrapper, which is the old
querySelector whenever there are fewer than two matches.

Discard cleanup now hides every duplicate wrapper rather than the first,
since a target inside a `.map()` renders one per item and hiding one left
the rest of the discarded variants on screen.

What still takes a raw first match is deliberate: bare existence checks,
selector strings for stylesheets and observers (which want to cover every
match), querySelectorAll sweeps, the parsed source document, and the
Svelte component wrapper, which holds no variant children at all. The
source-shape test pins that exact set by name, so a new raw lookup fails
until it is either routed through the resolver or justified there.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Live: a discard releases every wrapper it hid

Bugbot on #720: the non-restoreOriginal discard now hides every matching
wrapper, but the delayed fallback still released only the first
querySelector hit. A target inside a `.map()` renders one wrapper per
item, so the rest stayed at display:none and their original content never
came back on the static and missed-HMR flows that fallback exists for.

The hide, the existence checks, and the release now all speak about the
same set. discardedWrappers(sessionId) is the one place that collects it;
releaseDiscardedStaticWrappers takes the stylesheet down once and
releases each wrapper; releaseDiscardedStaticWrapper drops its sessionId
argument and just unwinds the node it is given. The HMR-ownership
decision still reads the first wrapper, which is fair: duplicates all
render from one source element, so ownership is uniform across them. The
reload branch is unchanged because a reload restores every original at
once.

Covered by a source-shape test rather than an e2e scenario:
hasFrameworkHmrOwnership is true for every React, Vue, and Svelte runtime
fixture, so all of them take the watcher path and none can reach the
static release. The existing framework-ownership guards in the same file
move to the new shape and keep their intent, including the one that says
only non-discard cleanup may blank the wrapper while waiting for HMR.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Release: publish the npm platform packages in one command

bun run release:platform-packages downloads each engine-v<ENGINE_VERSION>
binary with its .sha256 sidecar (required; nothing unverified is
published), stages the package from cli/platform-packages/<target> with
the version stamped, the executable at bin/ and the repo LICENSE, and
runs npm publish --access public. Targets already on the registry are
skipped so a re-run resumes after a partial failure. Preconditions:
package.json pins equal ENGINE_VERSION and npm is logged in.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* release-engine: pin checkout, upload-artifact and download-artifact at v7

The v4 pins target Node 20, which the runner now deprecates and forces
onto Node 24 with a warning on every step. The rest of the workflows
already use v7.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Tests: make the temp-dir helpers unique under a coarse clock

Windows' system clock is coarse enough that two parallel tests could get
the same pid-plus-nanoseconds directory name and then remove each
other's files (rust-windows: close_verb_round_trip_and_ownership,
NotFound). A per-process counter is appended to the name.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

* Tests: declare the temp-dir counter in the hook cache-root tests

The previous commit referenced TMP_SEQ there without defining it.

Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vau2X53xGTjjTCXWMVBoNY

---------

Co-authored-by: Claude <noreply@anthropic.com>
2026-09-04 10:42:45 -07:00

2352 lines
100 KiB
JavaScript

/**
* End-to-end live-mode tests — full click-to-accept cycle.
*
* For every framework fixture with a `runtime` block in fixture.json, this
* runner exercises the entire user-visible chain:
*
* 1. Stage → install → start live-server + dev server → inject script tag
* 2. Open Playwright Chromium, assert the live handshake fires
* 3. Spawn a deterministic fake-agent polling loop in this same process
* 4. Steer smoke: submit page-level chat → agent steer_done → bar unlocks
* 5. Drive the bar UI: pick element → Go → wait CYCLING → cycle → Accept
* 6. Assert source rewrite (variants block, then accepted-only after accept)
* 7. Assert DOM reflects the accepted variant via getComputedStyle
* 8. Tear down (browser, dev server, agent loop, live-server, tmp)
*
* The fake and LLM agents share one interface — see tests/live-e2e/agent.mjs
* and tests/live-e2e/agents/llm-agent.mjs.
*
* Run with: bun run test:live-e2e
*/
import { describe, it, before, after } from 'node:test';
import assert from 'node:assert/strict';
import { execFileSync } from 'node:child_process';
import { appendFileSync, cpSync, existsSync, mkdirSync, readdirSync, readFileSync, realpathSync, writeFileSync } from 'node:fs';
import { dirname, join, relative } from 'node:path';
import { fileURLToPath } from 'node:url';
import {
createFakeAgent,
republishSvelteComponentVariants,
FAKE_VARIANT_FONT_WEIGHTS,
} from './live-e2e/agent.mjs';
import { createLlmAgent, resolveLlmAgentConfig } from './live-e2e/agents/llm-agent.mjs';
import { bootFixtureSession, ENGINE_BIN, ENGINE_MISSING_MESSAGE, FIXTURES_DIR, runEngineSync } from './live-e2e/session.mjs';
import {
assertApplyDockVisible,
assertApplyDockLoading,
assertAnnotationUploadEvent,
assertSourceApplied,
chooseTuneStep,
clickExitLiveMode,
cycleToVariant,
clickAccept,
clickApplyEdits,
clickEditCopy,
clickDiscard,
clickMountRetry,
clickSaveEdit,
clickGo,
clickNext,
clickPrev,
editTextLeaf,
drawAnnotationPinAndStroke,
getVisibleVariant,
installLiveQueryHelpers,
isMountErrorCardVisible,
openTunePanel,
pickElement,
readComputedFontWeight,
runLiveChromeBottomBarSmoke,
setTuneRange,
waitForApplyDockHidden,
waitForBarHidden,
waitForComputedFontWeight,
waitForCycling,
waitForMountErrorCard,
waitForMountErrorCardGone,
runInsertFlow,
waitForHandshake,
} from './live-e2e/ui.mjs';
import { runSteerSmoke } from './live-e2e/steer.mjs';
import { runPreActions, waitForCyclingRobust } from './live-e2e/preactions.mjs';
const __dirname = dirname(fileURLToPath(import.meta.url));
// Discover fixtures that opt into the runtime E2E pass.
function listRuntimeFixtures() {
const names = readdirSync(FIXTURES_DIR, { withFileTypes: true })
.filter((e) => e.isDirectory())
.map((e) => e.name);
const out = [];
for (const name of names) {
const fixturePath = join(FIXTURES_DIR, name, 'fixture.json');
if (!existsSync(fixturePath)) continue;
const fixture = JSON.parse(readFileSync(fixturePath, 'utf-8'));
if (fixture.runtime) out.push({ name, fixture });
}
return out;
}
const allFixtures = listRuntimeFixtures();
// During development of the full-cycle test, a fixture subset is much faster
// to iterate on. Set IMPECCABLE_E2E_ONLY=<name>[,<name>...] to scope the run.
const onlyNames = parseFixtureFilter(process.env.IMPECCABLE_E2E_ONLY);
const fixtures = onlyNames.size > 0
? allFixtures.filter((f) => onlyNames.has(f.name))
: allFixtures;
const missingOnlyNames = [...onlyNames].filter((name) => !allFixtures.some((f) => f.name === name));
if (missingOnlyNames.length > 0) {
throw new Error(`Unknown IMPECCABLE_E2E_ONLY fixture(s): ${missingOnlyNames.join(', ')}`);
}
const manualOnly = process.env.IMPECCABLE_E2E_MANUAL_ONLY === '1'
|| process.env.IMPECCABLE_E2E_MANUAL_ONLY === 'true';
const reloadVariants = process.env.IMPECCABLE_E2E_RELOAD_VARIANTS === '1'
|| process.env.IMPECCABLE_E2E_RELOAD_VARIANTS === 'true';
const scenarioNames = parseFixtureFilter(process.env.IMPECCABLE_E2E_SCENARIOS);
const liveE2eTestTimeoutMs = readPositiveIntEnv('IMPECCABLE_E2E_TEST_TIMEOUT_MS');
const liveE2eTestOptions = liveE2eTestTimeoutMs ? { timeout: liveE2eTestTimeoutMs } : {};
// Widens the window between the server-side preflight scaffold (which triggers
// a framework HMR reload) and the agent's variant write. Used to reproduce
// races where the browser observes a wrapper with zero variants mid-generation.
const atomicDelayMs = readPositiveIntEnv('IMPECCABLE_E2E_ATOMIC_DELAY_MS') || 0;
if (fixtures.length === 0) {
describe('live-e2e (no runtime fixtures registered)', () => {
it('is a no-op', () => assert.ok(true));
});
}
let playwright;
let browser;
function parseFixtureFilter(value) {
return new Set(
String(value || '')
.split(/[,\s]+/)
.map((name) => name.trim())
.filter(Boolean),
);
}
function readPositiveIntEnv(name) {
const raw = process.env[name];
if (raw == null || raw === '') return null;
const parsed = Number(raw);
return Number.isFinite(parsed) && parsed > 0 ? parsed : null;
}
function shouldRunScenario(name) {
return scenarioNames.size === 0 || scenarioNames.has('all') || scenarioNames.has(name);
}
// Anything served out of the live preview / runtime tree. A 404 here is never
// benign: it means the variant preview module or its runtime never loaded, and
// the flow silently fell back to the untouched original.
const LIVE_TREE_URL_RE = /\.impeccable-live|impeccable\/live\/preview/i;
// The two framework dev-mode notices every React fixture emits.
const FRAMEWORK_NOISE_RE = /Download the React DevTools|StrictMode/i;
// Chromium's resource-failure console line. Only the favicon flavour is
// allowlisted; a favicon is not part of any fixture and its absence proves
// nothing about live mode.
const RESOURCE_404_RE = /Failed to load resource: the server responded with a status of 404/i;
const FAVICON_URL_RE = /favicon(\.[a-z0-9]+)?(\?|$)|\/favicon/i;
function isLiveTreeUrl(url) {
return LIVE_TREE_URL_RE.test(String(url || ''));
}
function isBenignConsoleError(entry) {
const text = String(entry || '');
// Live-tree failures are never allowlisted, whatever else they match.
if (LIVE_TREE_URL_RE.test(text)) return false;
if (FRAMEWORK_NOISE_RE.test(text)) return true;
if (RESOURCE_404_RE.test(text)) return FAVICON_URL_RE.test(text);
return false;
}
before(async () => {
if (fixtures.length === 0) return;
// The whole sweep drives engine verbs; without a binary there is nothing to
// test, and a silent skip would read as coverage.
if (!ENGINE_BIN) throw new Error(ENGINE_MISSING_MESSAGE);
try {
playwright = await import('playwright');
} catch (err) {
throw new Error(
`Playwright is required for live-e2e tests (${err.message}). Run: npx playwright install chromium`,
);
}
try {
browser = await launchLiveE2eBrowser();
} catch (err) {
throw new Error(`Failed to launch Chromium (${err.message}). Run: npx playwright install chromium`);
}
});
after(async () => {
if (browser) await browser.close();
});
async function launchLiveE2eBrowser() {
return playwright.chromium.launch({ headless: true });
}
async function teardownAndResetBrowser(teardown) {
try {
await teardown();
} finally {
if (browser) await browser.close().catch(() => {});
browser = await launchLiveE2eBrowser();
}
}
for (const { name, fixture } of fixtures) {
describe(`live-e2e · ${name} (${fixture.runtime.styling || 'unknown-styling'})`, () => {
it('drives the full click → Go → cycle → accept cycle', liveE2eTestOptions, async (t) => {
if (!shouldRunScenario('core')) {
t.skip('scenario filter excludes core');
return;
}
if (manualOnly || process.env.IMPECCABLE_E2E_MANUAL_SCENARIO) {
t.skip('manual scenario filter is active');
return;
}
// Fixtures may declare `runtime.knownLimitation` to flag a scenario
// that exposes a genuine live-mode gap rather than a test bug. The
// test still attempts the full chain but does not fail the suite when
// the documented failure mode appears — it surfaces the diagnostic so
// the limitation is visible in the run output.
const knownLimitation = fixture.runtime.knownLimitation;
// Pick the agent. `IMPECCABLE_E2E_AGENT=llm` opts into Claude first,
// with DeepSeek as the secondary fallback/override; everything else
// uses the deterministic fake. Skip rather than fail when LLM is
// requested but the selected provider key is missing so default suite
// runs in unauthenticated environments still pass.
const agentMode = process.env.IMPECCABLE_E2E_AGENT || 'fake';
let agent;
if (agentMode === 'llm') {
const llmConfig = resolveLlmAgentConfig({
model: process.env.IMPECCABLE_E2E_LLM_MODEL,
});
agent = await createLlmAgent({
config: llmConfig,
log: (m) => t.diagnostic('[llm] ' + m),
});
if (!agent) {
t.skip(`IMPECCABLE_E2E_AGENT=llm with provider=${llmConfig.provider} requires ${llmConfig.requiredEnv}`);
return;
}
t.diagnostic(`Using LLM agent (provider=${llmConfig.provider} model=${llmConfig.model})`);
} else {
agent = createFakeAgent();
}
t.diagnostic(`Booting fixture ${name}`);
const session = await bootFixtureSession({
name,
fixture,
browser,
agent,
wrapTarget: wrapTargetFromPickedElement,
atomicDelayMs,
log: (m) => t.diagnostic(m),
});
// `tmp` is the staged repo root; `appRoot` is the directory the dev
// server serves. They differ only for fixtures declaring
// `runtime.appDir`. Every fixture-relative source path resolves against
// appRoot — the repo root may not contain a single source file.
const { page, tmp, appRoot, consoleErrors, teardown } = session;
const expectedCount = 3;
const isInsert = fixture.runtime.mode === 'insert';
const insertCfg = fixture.runtime.insert || {};
const pickSelector = fixture.runtime.pickSelector || 'h1.hero-title';
const insertDomSelector = agentMode === 'llm' && insertCfg.expectSelectorLlm
? insertCfg.expectSelectorLlm
: (insertCfg.expectSelector || '.inserted-strip');
const domSelector = isInsert
? insertDomSelector
: pickSelector;
const usesSvelteComponentPreview = fixtureUsesSvelteKitAdapter(fixture) || name === 'nuxt-vite7';
// Component previews mount every variant into the same node, so one
// selector serves all three; the HTML/JSX paths keep a div per variant.
const variantContentSelectorFor = (variantNum) => (isInsert
? (usesSvelteComponentPreview ? '.inserted-copy' : `[data-impeccable-variant="${variantNum}"] .inserted-copy`)
: usesSvelteComponentPreview
? pickSelector
: `[data-impeccable-variant="${variantNum}"] > :first-child`);
let stateProbeBaseline = null;
let sourceFile = null;
try {
// 0. Root resolution — only for fixtures whose app is not the repo
// root. live.mjs booted from the repo root and had to find the app
// on its own; everything after this depends on it having done so.
if (session.appDir) {
t.diagnostic(`Asserting resolved roots for nested app ${session.appDir}/`);
assertNestedAppRoots(session);
}
// 1. Handshake
t.diagnostic('Waiting for live handshake');
await waitForHandshake(page);
if (fixture.runtime.liveChrome?.bottomBar) {
t.diagnostic('Running live chrome bottom-bar smoke');
await runLiveChromeBottomBarSmoke(page, {
expectDetectMinCount: fixture.runtime.liveChrome.detect?.expectMinCount || 1,
designTitle: fixture.runtime.liveChrome.design?.title || '',
designRawText: fixture.runtime.liveChrome.design?.rawText || '',
});
}
// 1b. Steer smoke — page-level chat before the heavier generate cycle.
if (fixture.runtime.steer !== false) {
const steerTimeouts = agentMode === 'llm'
? { unlockTimeoutMs: 90_000, selectorTimeoutMs: 45_000, runPreActions }
: { runPreActions };
await runSteerSmoke(page, appRoot, fixture, (m) => t.diagnostic(m), steerTimeouts);
}
// 2. preActions — fixtures with hidden/conditional content (modals,
// tabs, routes) drive the page into the right state before pick.
if (fixture.runtime.preActions) {
t.diagnostic(`Running ${fixture.runtime.preActions.length} preAction(s)`);
await runPreActions(page, fixture.runtime.preActions);
if (fixture.runtime.stateProbe) {
stateProbeBaseline = await assertStateProbe(page, fixture.runtime.stateProbe, 'after preActions');
}
}
// 3. Start generate — replace picks an element; insert places a placeholder.
if (isInsert) {
t.diagnostic(`Insert after ${insertCfg.anchorSelector || 'anchor'}`);
await runInsertFlow(page, {
anchorSelector: insertCfg.anchorSelector || 'section#features',
position: insertCfg.position || 'after',
prompt: insertCfg.prompt || 'Add new content',
});
} else {
t.diagnostic(`Picking ${pickSelector}`);
await pickElement(page, pickSelector, { position: fixture.runtime.pickPosition });
if (process.env.IMPECCABLE_E2E_DEBUG) {
const barText = await page.evaluate(() => {
const bar = document.querySelector('#impeccable-live-bar');
return bar ? { display: bar.style.display, text: bar.textContent || '', html: bar.innerHTML.slice(0, 500) } : null;
});
t.diagnostic(`Bar after pick: ${JSON.stringify(barText)}`);
}
t.diagnostic('Clicking Go');
await clickGo(page);
}
// 4. Wait for the agent's variants to land (HMR + MutationObserver).
// For fixtures whose picked element lives inside a conditional
// render (modal, tab, route), HMR can remount the parent and lose
// the open/active state — the wrapper exists in source but isn't
// in the DOM, so MutationObserver never sees it. Live mode now
// surfaces a toast asking the user to retrace the path; we mirror
// that here by re-running preActions on the first short timeout.
//
// The first-pass timeout has to be long enough to cover the agent's
// generate latency before declaring "state was lost, retrace." A
// fake agent finishes in <100ms. The real LLM path usually lands
// quickly too, but full-matrix runs can see minute-scale API or
// install pressure, so keep this gate patient enough that we do
// not retrace while the agent is still writing the variants.
t.diagnostic(`Waiting for CYCLING state with ${expectedCount} variants`);
await waitForCyclingRobust(page, expectedCount, {
agentMode,
preActions: fixture.runtime.preActions,
log: (m) => t.diagnostic(m),
});
if (fixture.runtime.stateProbe) {
await assertStateProbe(page, fixture.runtime.stateProbe, 'after variants', { baseline: stateProbeBaseline });
}
// 5. Source-side check: wrapper + style + variants are present
sourceFile = await locateSessionFile(appRoot);
if (session.appDir) {
assert.ok(
relative(tmp, sourceFile).startsWith(session.appDir + '/'),
`session source must live under ${session.appDir}/, got ${relative(tmp, sourceFile)}`,
);
}
const after = readFileSync(sourceFile, 'utf-8');
const svelteComponentSession = svelteComponentTargetFor(sourceFile);
if (svelteComponentSession) {
const componentExtension = svelteComponentSession.manifest.componentExtension || 'svelte';
const variantFile = join(appRoot, svelteComponentSession.manifest.componentDir, `v2.${componentExtension}`);
const variantBody = readFileSync(variantFile, 'utf-8');
const routeBody = readFileSync(join(appRoot, svelteComponentSession.manifest.sourceFile), 'utf-8');
assert.match(after, /"previewMode": "(?:svelte|vue)-component"/, 'framework component manifest inserted');
if (isInsert) {
assert.equal(svelteComponentSession.manifest.mode, 'insert', 'Svelte insert manifest marks insert mode');
if (agentMode === 'fake') {
assert.match(variantBody, /inserted-strip/, 'Svelte insert variant component contains inserted content');
} else if (insertCfg.expectSourcePattern) {
assert.match(variantBody, new RegExp(insertCfg.expectSourcePattern, 'i'), 'Svelte insert variant component contains prompt-matching content');
} else {
assert.match(variantBody, /<([a-z][\w:-]*)\b[\s\S]*<\/\1>|<[a-z][\w:-]*\b[^>]*\/>/i, 'Svelte insert variant component contains a root element');
}
} else {
assert.match(variantBody, new RegExp(`<${svelteComponentSession.expectedTag}\\b`), 'component variant contains target element');
}
assert.doesNotMatch(routeBody, /data-impeccable-variants="/, 'route source is not edited during component preview');
} else {
assert.match(after, /data-impeccable-variants="/, 'wrapper inserted');
}
if (isInsert) {
if (svelteComponentSession) {
assert.equal(svelteComponentSession.manifest.mode, 'insert', 'Svelte insert uses component preview mode');
} else {
assert.match(after, /data-impeccable-mode="insert"/, 'insert mode wrapper');
assert.doesNotMatch(after, /data-impeccable-variant="original"/, 'insert has no original variant');
}
if (insertCfg.assertAnchorContains) {
const anchorSource = svelteComponentSession
? readFileSync(join(appRoot, svelteComponentSession.manifest.sourceFile), 'utf-8')
: after;
assert.match(anchorSource, new RegExp(insertCfg.assertAnchorContains), 'anchor section untouched');
}
}
if (svelteComponentSession) {
const componentExtension = svelteComponentSession.manifest.componentExtension || 'svelte';
assert.match(readFileSync(join(appRoot, svelteComponentSession.manifest.componentDir, `v2.${componentExtension}`), 'utf-8'), /<style\b/, 'component variant has a style block');
} else if (sourceFile.endsWith('.astro')) {
assert.match(after, /<style is:inline data-impeccable-css="/, 'Astro live CSS uses an inline compiler-bypassing style block');
assert.match(
after,
/\[data-impeccable-variant="1"\]\s*>\s*(?:h1|\.[\w-]+)/,
'event=live_e2e.astro_css_prefix actor=agent operation=write_variants risk=astro_scopes_preview_css_away expected=variant-prefixed global selector actual=missing suggestion=inspect fake agent styleMode handling',
);
assert.doesNotMatch(after, /@scope \(\[data-impeccable-variant="1"\]\)/, 'Astro live CSS does not use raw @scope');
} else {
assert.match(after, /<style data-impeccable-css="/, 'colocated <style> block present');
assert.match(after, /@scope \(\[data-impeccable-variant="1"\]\)/, 'scoped CSS for variant 1');
assert.match(after, /@scope \(\[data-impeccable-variant="2"\]\)/, 'scoped CSS for variant 2');
assert.match(after, /@scope \(\[data-impeccable-variant="3"\]\)/, 'scoped CSS for variant 3');
}
// Param manifest assertions are scoped to fake-agent mode. The fake
// agent deterministically emits one param per variant covering all
// three kinds; the LLM agent is non-deterministic and may legitimately
// emit no params per the live.md spec ("variants are fixed points").
if (agentMode === 'fake') {
const paramsSource = svelteComponentSession
? readFileSync(join(appRoot, svelteComponentSession.manifest.componentDir, 'params.json'), 'utf-8')
: after;
assert.match(paramsSource, svelteComponentSession ? /"1"\s*:/ : /data-impeccable-params=/, 'params manifest emitted');
for (const kind of ['range', 'steps', 'toggle']) {
assert.match(paramsSource, new RegExp(`"kind"\\s*:\\s*"${kind}"`), `param kind ${kind} present`);
}
await page.waitForFunction(() => {
const root = window.__IMPECCABLE_LIVE_CHROME_CORE__?.root?.()
|| window.__IMPECCABLE_LIVE_UI_ROOT__
|| document;
const tune = root.querySelector('[data-iceq-tune="1"]');
return tune && tune.disabled === false && /Tune/.test(tune.textContent || '');
}, { timeout: 5_000 });
}
// 6. Cycle variants. Most fixtures stop at variant 2; Svelte Insert
// also exercises right/right/left/right and accepts variant 3.
const cycleSequence = Array.isArray(fixture.runtime.variantSequence) && fixture.runtime.variantSequence.length > 0
? fixture.runtime.variantSequence
: [2];
let visible = await readVisibleVariantForCycle(page);
// Render proof, not bar-label proof: the fake agent gives every variant
// a distinct font-weight, so a computed read says which variant the
// user is actually looking at. Checked for each variant the run reaches.
const styleCheckedVariants = new Set();
const assertVisibleVariantStyle = async (variantNum) => {
if (agentMode !== 'fake') return;
const expected = FAKE_VARIANT_FONT_WEIGHTS[variantNum];
if (!expected || styleCheckedVariants.has(variantNum)) return;
styleCheckedVariants.add(variantNum);
const selector = variantContentSelectorFor(variantNum);
await waitForComputedFontWeight(page, selector, expected, { timeout: 5_000 }).catch(() => {});
const variantWeight = await readComputedFontWeight(page, selector).catch(() => null);
if (variantWeight !== expected) {
const styleSnapshot = await evaluatePageWithTimeout(
page,
(sel) => {
const query = window.__impeccableLiveQuery || ((s) => document.querySelector(s));
const el = query(sel) || document.querySelector(sel);
const styleEl = document.querySelector('style[data-impeccable-css]');
const rules = [];
for (const sheet of [...document.styleSheets]) {
if (sheet.ownerNode !== styleEl) continue;
try {
rules.push(...[...sheet.cssRules].map((rule) => rule.cssText));
} catch (err) {
rules.push(`cssRules error: ${err.message}`);
}
}
return {
selector: sel,
element: el?.outerHTML || null,
parent: el?.parentElement?.outerHTML?.slice(0, 800) || null,
computedWeight: el ? getComputedStyle(el).fontWeight : null,
styleText: styleEl?.textContent || null,
rules,
};
},
selector,
5_000,
'variant style snapshot',
).catch((err) => ({ error: err.message }));
t.diagnostic(`--- variant ${variantNum} style snapshot ---`);
t.diagnostic(JSON.stringify(styleSnapshot, null, 2));
}
assert.equal(
variantWeight,
expected,
`event=live_e2e.variant_css_applied actor=browser operation=render_visible_variant variant=${variantNum} risk=unstyled_live_preview expected=font-weight ${expected} actual=${variantWeight} suggestion=inspect live CSS style mode and selector shape`,
);
};
await assertVisibleVariantStyle(visible);
// Optional fixture hook: assert attribute/text values inside the
// MOUNTED variant DOM. Component previews hydrate collection items
// from the rendered page (text slots and attribute slots); a probe
// here catches a preview that mounts but with empty hydrated values,
// which every other assertion (style, counter, accept) misses.
if (Array.isArray(fixture.runtime.mountedDomProbe)) {
for (const probe of fixture.runtime.mountedDomProbe) {
const actual = await evaluatePageWithTimeout(
page,
({ sel, attr }) => {
const query = window.__impeccableLiveQuery || ((s) => document.querySelector(s));
const el = query(sel) || document.querySelector(sel);
if (!el) return null;
return attr ? el.getAttribute(attr) : (el.textContent || '').trim();
},
{ sel: probe.selector, attr: probe.attr || null },
5_000,
'mounted DOM probe',
);
assert.equal(
actual,
probe.expect,
`mounted variant DOM: ${probe.selector}${probe.attr ? ` [${probe.attr}]` : ' text'}`,
);
}
}
for (const targetVariant of cycleSequence) {
t.diagnostic(`Cycling to variant ${targetVariant}`);
visible = await cycleToVariant(page, targetVariant, expectedCount, {
settleTimeout: agentMode === 'llm' ? 60_000 : 15_000,
});
assert.equal(visible, targetVariant, `variant ${targetVariant} visible`);
await assertVisibleVariantStyle(targetVariant);
}
if (reloadVariants && usesSvelteComponentPreview) {
const visibleBeforeReload = await getVisibleVariant(page);
t.diagnostic(`Reload recovery probe at variant ${visibleBeforeReload}/${expectedCount}`);
const savedBeforeReload = await readLiveSessionStorage(page);
assert.ok(savedBeforeReload, 'local session exists before reload');
assert.equal(savedBeforeReload.visible, visibleBeforeReload, 'local session stores visible variant before reload');
assert.equal(savedBeforeReload.previewMode, 'svelte-component', 'local session stores Svelte preview mode before reload');
assert.ok(savedBeforeReload.previewFile, 'local session stores Svelte preview manifest before reload');
await page.reload({ waitUntil: 'domcontentloaded' });
await waitForHandshake(page);
const savedAfterReload = await readLiveSessionStorage(page);
assert.ok(savedAfterReload, 'local session exists after reload');
assert.equal(savedAfterReload.id, savedBeforeReload.id, 'same live session id survives refresh');
assert.equal(savedAfterReload.visible, visibleBeforeReload, 'fresh local visible variant wins after refresh');
assert.equal(savedAfterReload.previewFile, savedBeforeReload.previewFile, 'preview manifest survives refresh');
if (fixture.runtime.preActions?.length) {
await waitForRecoverableVariantSession(page, visibleBeforeReload, expectedCount, {
timeout: agentMode === 'llm' ? 60_000 : 15_000,
});
await runPreActions(page, fixture.runtime.preActions);
}
await waitForCyclingRobust(page, expectedCount, {
agentMode,
preActions: fixture.runtime.preActions,
log: (m) => t.diagnostic(m),
});
await waitForVariantCounter(page, visibleBeforeReload, expectedCount, {
timeout: agentMode === 'llm' ? 60_000 : 15_000,
});
assert.equal(await getVisibleVariant(page), visibleBeforeReload, 'same visible variant is restored after refresh');
if (visibleBeforeReload < expectedCount) {
await clickNext(page);
await waitForVariantCounter(page, visibleBeforeReload + 1, expectedCount, {
timeout: agentMode === 'llm' ? 60_000 : 15_000,
});
assert.equal(await getVisibleVariant(page), visibleBeforeReload + 1, 'next arrow still works after refresh restore');
await clickPrev(page);
await waitForVariantCounter(page, visibleBeforeReload, expectedCount, {
timeout: agentMode === 'llm' ? 60_000 : 15_000,
});
assert.equal(await getVisibleVariant(page), visibleBeforeReload, 'prev arrow still works after refresh restore');
} else {
await clickPrev(page);
await waitForVariantCounter(page, visibleBeforeReload - 1, expectedCount, {
timeout: agentMode === 'llm' ? 60_000 : 15_000,
});
assert.equal(await getVisibleVariant(page), visibleBeforeReload - 1, 'prev arrow still works after refresh restore');
await clickNext(page);
await waitForVariantCounter(page, visibleBeforeReload, expectedCount, {
timeout: agentMode === 'llm' ? 60_000 : 15_000,
});
assert.equal(await getVisibleVariant(page), visibleBeforeReload, 'next arrow still works after refresh restore');
}
}
// 7. Accept the final visible variant
const acceptVariant = cycleSequence[cycleSequence.length - 1] || 2;
t.diagnostic(`Accepting variant ${acceptVariant}`);
await clickAccept(page, { expectedVariant: acceptVariant });
await waitForBarHidden(page);
const sourceShadow = !!sourceShadowTargetFor(sourceFile);
const svelteComponentTarget = svelteComponentSession || svelteComponentTargetFor(sourceFile);
const svelteComponent = !!svelteComponentTarget;
if (fixture.runtime.stateProbe && !svelteComponent) {
await assertStateProbe(page, fixture.runtime.stateProbe, 'after accept', { baseline: stateProbeBaseline });
}
if (sourceShadow && typeof session.stopLiveServer === 'function') {
t.diagnostic('Stopping live-server to flush deferred accept');
session.stopLiveServer();
}
// 8. Wait for live-accept + the agent's carbonize cleanup to land.
// File-side: wrapper, all variants, and carbonize markers gone;
// only the accepted inner element survives.
t.diagnostic('Waiting for accept + carbonize cleanup to land');
const final = await waitForSourceClean(sourceFile, 20_000, { svelteComponentTarget });
if (svelteComponentTarget) {
assert.equal(existsSync(svelteComponentTarget.manifestPath), false, 'Svelte temp preview session removed after accept');
const snapshotPath = join(appRoot, '.impeccable/live/sessions', `${svelteComponentTarget.manifest.id}.snapshot.json`);
const snapshot = JSON.parse(readFileSync(snapshotPath, 'utf-8'));
assert.equal(snapshot.phase, 'completed');
assert.equal(snapshot.sourceFile, svelteComponentTarget.manifest.sourceFile);
assert.doesNotMatch(snapshot.sourceFile, /node_modules\/\.impeccable-live|\.impeccable\/live\/previews/);
}
assert.doesNotMatch(final, /data-impeccable-variants="/, 'variants wrapper removed');
assert.doesNotMatch(final, /impeccable-variants-start/, 'variants-start marker removed');
assert.doesNotMatch(final, /impeccable-carbonize-start/, 'carbonize-start marker removed');
assert.doesNotMatch(final, /impeccable-carbonize-end/, 'carbonize-end marker removed');
assert.doesNotMatch(final, /data-impeccable-carbonize=/, 'carbonize wrapper removed');
assert.doesNotMatch(final, /data-impeccable-variant="/, 'no leftover variant scaffolding');
if (isInsert) {
if (agentMode === 'fake') {
assert.match(final, /inserted-strip/, 'accepted insert content survives');
} else if (insertCfg.expectSourcePattern) {
assert.match(final, new RegExp(insertCfg.expectSourcePattern, 'i'), 'accepted insert content survives');
}
if (insertCfg.assertAnchorContains) {
assert.match(final, new RegExp(insertCfg.assertAnchorContains), 'anchor section still in source');
}
} else {
const acceptedSourcePattern = fixture.runtime.acceptedSourcePattern
|| '<h1[^>]*(class|className)="[^"]*\\bhero-title\\b[^"]*"';
assert.match(
final,
new RegExp(acceptedSourcePattern),
'accepted source element survives',
);
}
// Optional fixture hook: assert that arbitrary strings survive the
// wrap → accept → carbonize cycle. Used by repeated-branch fixtures
// to prove wrap disambiguated correctly — sibling branches the test
// didn't pick should be untouched.
if (Array.isArray(fixture.runtime.assertSourceContains)) {
for (const needle of fixture.runtime.assertSourceContains) {
assert.ok(
final.includes(needle),
`source still contains ${JSON.stringify(needle)} after accept (sibling branch must not be rewritten)`,
);
}
}
// 9. DOM-side: at least one matching element, none inside any wrapper.
if (svelteComponent && fixture.runtime.preActions) {
await runPreActions(page, fixture.runtime.preActions);
}
try {
await waitForAcceptedDom(page, domSelector, { allowVariantRoot: sourceShadow, timeout: 20_000 });
} catch (err) {
if (!svelteComponent || !fixture.runtime.preActions) throw err;
t.diagnostic('Accepted Svelte DOM was not visible after HMR; reloading and re-running preActions');
await page.reload({ waitUntil: 'domcontentloaded' });
await waitForHandshake(page);
await runPreActions(page, fixture.runtime.preActions);
await waitForAcceptedDom(page, domSelector, { allowVariantRoot: sourceShadow, timeout: 20_000 });
}
// 9b. reloadProbe — fixtures with conditional render assert that the
// accepted variant survives a full page reload. The picked element
// may be hidden by default (closed modal, non-default tab); the
// probe re-runs preActions to bring it back into the DOM.
if (fixture.runtime.reloadProbe) {
t.diagnostic('Running reloadProbe (reload + reach + assert)');
await page.reload({ waitUntil: 'domcontentloaded' });
if (fixture.runtime.reloadProbe.preActions) {
await runPreActions(page, fixture.runtime.reloadProbe.preActions);
}
const expectSelector = fixture.runtime.reloadProbe.expectSelector || pickSelector;
await page.waitForSelector(expectSelector, { timeout: 10_000 });
}
// 9c. Network hygiene — nothing under the live preview/runtime tree
// may 404 or fail. A missing `.impeccable-live` module means the
// preview never mounted, which the DOM assertions above can miss
// when the original element still renders.
const liveTreeFailures = (session.failedRequests || []).filter((f) => isLiveTreeUrl(f.url));
assert.equal(
liveTreeFailures.length,
0,
`live preview/runtime requests must all succeed, got:\n${
liveTreeFailures.map((f) => `${f.reason} ${f.url}`).join('\n')
}`,
);
// 10. Console hygiene — no errors during the whole flow.
if (fixture.runtime.probe?.expectConsoleClean) {
const realErrors = consoleErrors.filter((e) => !isBenignConsoleError(e));
if (realErrors.length > 0) {
t.diagnostic('--- console errors ---');
for (const e of realErrors) t.diagnostic(e);
t.diagnostic('--- final source ---');
t.diagnostic(readFileSync(sourceFile, 'utf-8'));
}
assert.equal(
realErrors.length,
0,
`expected clean console, got:\n${realErrors.join('\n')}`,
);
}
} catch (err) {
await captureLiveE2eFailure({
name,
fixture,
session,
sourceFile,
error: err,
log: (m) => t.diagnostic(m),
});
if (knownLimitation) {
t.diagnostic(`KNOWN LIMITATION: ${knownLimitation}`);
t.diagnostic(`Failure: ${err.message?.split('\n')[0] || err}`);
t.skip(`known limitation: ${knownLimitation}`);
return;
}
throw err;
} finally {
await teardownAndResetBrowser(teardown);
}
});
// -----------------------------------------------------------------
// Params end-to-end: knob → accept → baked literal in source.
//
// The Svelte accept pipeline bakes params mechanically, so the unit
// tests already cover the transform. What they cannot cover is that the
// value the user actually dialled in the Tune popover is the value that
// reaches it. This drives the real controls and reads the real source.
// -----------------------------------------------------------------
if (shouldRunScenario('params') && fixture.runtime.paramsScenario) {
it('bakes tuned param values into the accepted source', liveE2eTestOptions, async (t) => {
if (!canRunFakeAgentScenario(t)) return;
const scenario = fixture.runtime.paramsScenario;
const targetVariant = scenario.variant || 2;
const run = await bootComponentScenario({ t, name, fixture, expectedCount: 3 });
const { session, sourceFile, svelteComponentTarget } = run;
const { page } = session;
try {
await cycleToVariant(page, targetVariant, 3);
t.diagnostic('Opening the Tune popover');
await openTunePanel(page);
const applied = await setTuneRange(page, scenario.rangeLabel, scenario.rangeValue);
assert.equal(applied, scenario.rangeValue, 'range knob took the requested value');
await chooseTuneStep(page, scenario.stepsLabel, scenario.stepsOptionLabel);
t.diagnostic(`Accepting variant ${targetVariant} with tuned params`);
await clickAccept(page, { expectedVariant: targetVariant });
await waitForBarHidden(page);
const final = await waitForSourceClean(sourceFile, 20_000, { svelteComponentTarget });
for (const needle of scenario.expectSourceContains || []) {
assert.ok(
final.includes(needle),
`accepted source should bake ${JSON.stringify(needle)}; got:\n${final}`,
);
}
for (const needle of scenario.expectSourceMissing || []) {
assert.equal(
final.includes(needle),
false,
`accepted source should drop the unchosen branch ${JSON.stringify(needle)}; got:\n${final}`,
);
}
// The generic contract, independent of this fixture's values: every
// preview-only param hook is gone once the source is accepted.
assert.doesNotMatch(final, /data-p-[A-Za-z0-9_-]+/, 'no data-p-* param attributes survive accept');
assert.doesNotMatch(final, /var\(--p-/, 'no unbaked var(--p-*) survives accept');
} finally {
await teardownAndResetBrowser(session.teardown);
}
});
}
// -----------------------------------------------------------------
// Failure injection for component previews.
// -----------------------------------------------------------------
if (fixture.runtime.componentFailureScenarios) {
const failure = fixture.runtime.componentFailureScenarios;
const brokenVariant = failure.variant || 2;
if (shouldRunScenario('mount-failure')) {
it('surfaces a broken variant without losing the session, and recovers on Retry', liveE2eTestOptions, async (t) => {
if (!canRunFakeAgentScenario(t)) return;
// Auto-repair off: this scenario is about what the USER sees and can
// do when a publish is unrenderable. An agent that silently
// republishes would hide exactly the surface under test.
const run = await bootComponentScenario({
t,
name,
fixture,
expectedCount: 3,
agent: createFakeAgent({ autoRepairMountFailures: false }),
});
const { session, svelteComponentTarget } = run;
const { page, appRoot } = session;
try {
const sessionId = svelteComponentTarget.manifest.id;
const variantPath = revisionVariantPath(appRoot, run.manifestPath, brokenVariant);
const saved = readFileSync(variantPath, 'utf-8');
// Corrupt rather than delete: a published variant that does not
// compile is the failure an agent actually produces, and it keeps
// the module resolvable so the recovery is about the content.
t.diagnostic(`Corrupting published revision file ${relative(appRoot, variantPath)}`);
writeFileSync(variantPath, '<div class="impeccable-broken-variant">{ not valid svelte\n', 'utf-8');
// One step forward onto the variant whose module is gone. The step
// deliberately does not settle, so drive the click directly.
t.diagnostic(`Stepping onto the broken variant ${brokenVariant}`);
await clickNext(page);
const cardText = await waitForMountErrorCard(page, { variant: brokenVariant });
assert.match(cardText, /Retry/, 'the mount-error card offers a Retry');
// The session must survive the failure: the old behaviour wiped
// local state and orphaned a session the server still tracked.
const barVisible = await page.evaluate(() => {
const bar = window.__impeccableLiveQuery('#impeccable-live-bar');
return Boolean(bar) && bar.style.display !== 'none';
});
assert.equal(barVisible, true, 'the cycling bar survives a mount failure');
const stored = await readLiveSessionStorage(page);
assert.equal(stored?.id, sessionId, 'the local session record survives a mount failure');
const events = await waitForJournalEvent(appRoot, sessionId, 'variant_mount_failed', 15_000);
const failedEvent = events.at(-1);
assert.equal(failedEvent.variant, brokenVariant, 'the journal names the variant that failed');
assert.ok(failedEvent.error, 'the journal carries the mount error text');
// A failed step reverts to the variant that still renders, so the
// comparison is usable rather than blank; Retry re-imports the
// session, and the repaired variant is reachable again.
t.diagnostic('Restoring the revision file and clicking Retry');
writeFileSync(variantPath, saved, 'utf-8');
await clickMountRetry(page);
await waitForMountErrorCardGone(page);
await cycleToVariant(page, brokenVariant, 3, { settleTimeout: 20_000 });
await waitForComputedFontWeight(
page,
fixture.runtime.pickSelector || 'h1.hero-title',
FAKE_VARIANT_FONT_WEIGHTS[brokenVariant],
{ timeout: 20_000 },
);
assert.equal(await getVisibleVariant(page), brokenVariant, 'the repaired variant is the visible one');
assert.equal(await isMountErrorCardVisible(page), false, 'the repaired mount raises no new error');
} finally {
await teardownAndResetBrowser(session.teardown);
}
});
}
if (shouldRunScenario('republish')) {
it('mounts republished variant content instead of a cached compile', liveE2eTestOptions, async (t) => {
if (!canRunFakeAgentScenario(t)) return;
const run = await bootComponentScenario({ t, name, fixture, expectedCount: 3 });
const { session } = run;
const { page, appRoot } = session;
const pickSelector = fixture.runtime.pickSelector || 'h1.hero-title';
try {
await cycleToVariant(page, brokenVariant, 3);
await waitForComputedFontWeight(page, pickSelector, FAKE_VARIANT_FONT_WEIGHTS[brokenVariant]);
const beforeRevision = readManifest(run.manifestPath).revision;
t.diagnostic('Republishing every variant with new styling');
await republishSvelteComponentVariants({
tmp: appRoot,
manifestFile: relative(appRoot, run.manifestPath),
live: session.live,
css: (variantNum, shape) => `${shape.rootSelector} { font-weight: ${100 * variantNum}; }`,
});
// A stale transform cache would keep serving the previous compile;
// the server-stamped revision dir is what makes the import path new.
await waitForComputedFontWeight(page, pickSelector, String(100 * brokenVariant), { timeout: 20_000 });
const afterRevision = readManifest(run.manifestPath).revision;
assert.ok(
afterRevision > beforeRevision,
`republish must bump the preview revision (${beforeRevision}${afterRevision})`,
);
assert.equal(await isMountErrorCardVisible(page), false, 'a clean republish shows no mount error');
} finally {
await teardownAndResetBrowser(session.teardown);
}
});
}
if (shouldRunScenario('storage-loss') && failure.storageLoss !== false) {
it('rehydrates the comparison from the server after localStorage is cleared', liveE2eTestOptions, async (t) => {
if (!canRunFakeAgentScenario(t)) return;
const run = await bootComponentScenario({ t, name, fixture, expectedCount: 3 });
const { session } = run;
const { page } = session;
try {
await cycleToVariant(page, brokenVariant, 3);
assert.ok(await readLiveSessionStorage(page), 'a local session record exists before the wipe');
t.diagnostic('Clearing localStorage and reloading');
await page.evaluate(() => localStorage.clear());
assert.equal(await readLiveSessionStorage(page), null, 'the local session record really was wiped');
await page.reload({ waitUntil: 'domcontentloaded' });
await waitForHandshake(page);
if (fixture.runtime.preActions) await runPreActions(page, fixture.runtime.preActions);
// Nothing local is left: reaching CYCLING again can only come from
// the server's durable session record.
await waitForCycling(page, 3, { timeout: 30_000 });
const restored = await readLiveSessionStorage(page);
assert.equal(restored?.id, run.svelteComponentTarget.manifest.id, 'the adopted session keeps its id');
assert.equal(restored?.expected, 3, 'the adopted session knows the variant count');
} finally {
await teardownAndResetBrowser(session.teardown);
}
});
}
}
if (shouldRunScenario('missed-done') && fixture.runtime.missedDoneReloadScenario) {
it('recovers when the preflight reload makes the browser miss the done broadcast', liveE2eTestOptions, async (t) => {
if (manualOnly || process.env.IMPECCABLE_E2E_MANUAL_SCENARIO) {
t.skip('manual scenario filter is active');
return;
}
const scenarioLimitation = fixture.runtime.missedDoneReloadScenario.knownLimitation;
if (scenarioLimitation) {
t.diagnostic(`KNOWN LIMITATION: ${scenarioLimitation}`);
t.skip(`known limitation: ${scenarioLimitation}`);
return;
}
// Deterministic reproduction of the race the CI astro-vite7 timeout
// exposed: the server-side preflight scaffold write triggers a
// framework full-reload, and the agent's variant write + `done` SSE
// land while the browser is mid-reload. The resumed page misses both
// signals and used to sit in GENERATING at 0/N forever. Here the
// reloaded page's HMR websocket connects to a dead mock and its SSE
// stream is refused until after the agent finishes, forcing the miss
// every time; recovery must come from the live server redelivering
// the completion once the browser's SSE reconnects.
const agent = createFakeAgent();
const session = await bootFixtureSession({
name,
fixture,
browser,
agent,
wrapTarget: wrapTargetFromPickedElement,
atomicDelayMs: 2500,
log: (m) => t.diagnostic(m),
});
const { page, appRoot, teardown } = session;
const pickSelector = fixture.runtime.pickSelector || 'h1.hero-title';
try {
await waitForHandshake(page);
// Arm the blocks before Go. They only affect connections opened
// after this point — the current page keeps its live SSE and HMR
// socket, so the scaffold-triggered reload still goes through; it is
// the page that boots FROM that reload which comes up deaf.
let liveConnectionsBlocked = true;
await page.context().route('**/events?token=*', (route) => {
if (liveConnectionsBlocked && route.request().method() === 'GET') {
return route.abort();
}
return route.continue();
});
await page.context().routeWebSocket('**', () => {
// Never connectToServer(): the reloaded page's HMR client talks to
// a dead mock, so the variant-write full-reload push is lost.
});
await pickElement(page, pickSelector);
t.diagnostic('Clicking Go (agent write delayed 2.5s; reloaded page will miss HMR + SSE)');
await clickGo(page);
// The scaffold write reloads the page into wrapped-but-empty source.
await page.waitForFunction(() => {
const wrapper = document.querySelector('[data-impeccable-variants]');
if (!wrapper) return false;
const variants = wrapper.querySelectorAll('[data-impeccable-variant]:not([data-impeccable-variant="original"])');
return variants.length === 0;
}, { timeout: 15_000 });
t.diagnostic('Browser reloaded into the scaffold-only wrapper (0 variants)');
// Capture the scaffold-only source now (the agent write is still
// ~1.5s away). The first post-reconnect /source read will be served
// this stale copy, forcing the completion-driven fallback through
// its retry path — a single no-retry read here strands the tab in
// GENERATING forever.
const scaffoldSourceFile = join(appRoot, fixture.runtime.missedDoneReloadScenario.sourceFile);
const scaffoldOnlySource = readFileSync(scaffoldSourceFile, 'utf-8');
let staleSourceServed = false;
await page.context().route('**/source?token=*', (route) => {
if (!staleSourceServed) {
staleSourceServed = true;
return route.fulfill({
status: 200,
contentType: 'text/html; charset=utf-8',
body: scaffoldOnlySource,
});
}
return route.continue();
});
// Wait for the delayed agent write to land in source while the page
// is deaf to both delivery channels.
const writeDeadline = Date.now() + 20_000;
for (;;) {
if (/data-impeccable-variant="3"/.test(readFileSync(scaffoldSourceFile, 'utf-8'))) break;
if (Date.now() > writeDeadline) throw new Error('agent variant write never landed in source');
await new Promise((resolve) => setTimeout(resolve, 200));
}
t.diagnostic('Variants in source; the browser missed the broadcast. Unblocking SSE.');
liveConnectionsBlocked = false;
// EventSource auto-retries. On reconnect the live server must
// redeliver the missed completion; the browser's recovery read gets
// the stale scaffold first, retries, and then injects the real
// variants from source — reaching CYCLING despite the dead HMR.
await waitForCycling(page, 3, { timeout: 30_000 });
assert.equal(staleSourceServed, true, 'the stale /source intercept must have exercised the retry path');
t.diagnostic('Session recovered to CYCLING after SSE redelivery + stale-read retry');
} finally {
await teardownAndResetBrowser(teardown);
}
});
}
if (shouldRunScenario('orphan') && fixture.runtime.orphanedWrapperScenario) {
it('self-discards an orphaned session when its wrapper is edited out of source', liveE2eTestOptions, async (t) => {
if (manualOnly || process.env.IMPECCABLE_E2E_MANUAL_SCENARIO) {
t.skip('manual scenario filter is active');
return;
}
// Repro of issue #439: a cycling session is abandoned (no Accept or
// Discard), the wrapped region is edited out of source, and the page
// reloads. The resumed session used to freeze the picker forever;
// recovery required a manual live-complete --discarded. It must now
// self-discard and hand the surface back to the picker.
const agent = createFakeAgent();
const session = await bootFixtureSession({
name,
fixture,
browser,
agent,
wrapTarget: wrapTargetFromPickedElement,
log: (m) => t.diagnostic(m),
});
const { page, appRoot, teardown } = session;
const cfg = fixture.runtime.orphanedWrapperScenario;
const pickSelector = fixture.runtime.pickSelector || 'h1.hero-title';
try {
await waitForHandshake(page);
const sourceFile = join(appRoot, cfg.sourceFile);
const pristine = readFileSync(sourceFile, 'utf-8');
await pickElement(page, pickSelector, { position: fixture.runtime.pickPosition });
await clickGo(page);
await waitForCyclingRobust(page, 3, { timeout: 60_000, log: (m) => t.diagnostic(m) });
const saved = await readLiveSessionStorage(page);
assert.ok(saved?.id, 'cycling session persisted to local storage');
t.diagnostic('Restoring pristine source (simulated external edit that removes the wrapper)');
writeFileSync(sourceFile, pristine);
await page.reload({ waitUntil: 'domcontentloaded' });
await waitForHandshake(page);
// Resume adopts the cycling session, the source read finds no
// wrapper, retries, then self-discards: local session cleared.
const deadline = Date.now() + 30_000;
for (;;) {
const current = await readLiveSessionStorage(page);
if (!current?.id) break;
if (Date.now() > deadline) throw new Error('orphaned session was never discarded');
await new Promise((resolve) => setTimeout(resolve, 500));
}
t.diagnostic('Orphaned session discarded; verifying durable phase + picker rearm');
const snapshotPath = join(appRoot, '.impeccable/live/sessions', `${saved.id}.snapshot.json`);
const snapshot = JSON.parse(readFileSync(snapshotPath, 'utf-8'));
assert.equal(
snapshot.phase,
'discarded',
'an orphaned discard is terminalized server-side without agent involvement',
);
// The regression that mattered: the picker must arm again.
await pickElement(page, pickSelector, { position: fixture.runtime.pickPosition });
} finally {
await teardownAndResetBrowser(teardown);
}
});
}
if (shouldRunScenario('foreign') && fixture.runtime.foreignSessionScenario) {
it('clears another project\'s leftover browser session instead of resuming it', liveE2eTestOptions, async (t) => {
if (manualOnly || process.env.IMPECCABLE_E2E_MANUAL_SCENARIO) {
t.skip('manual scenario filter is active');
return;
}
// Repro of the cross-project leak: localStorage is per-ORIGIN, and two
// projects routinely reuse the same localhost port across time. A
// leftover cycling session from the other project used to be resumed
// ("Variants ready" with no user action), and its checkpoints
// materialized a ghost session in THIS project's durable store that
// kept reattaching after every discard.
const agent = createFakeAgent();
const session = await bootFixtureSession({
name,
fixture,
browser,
agent,
wrapTarget: wrapTargetFromPickedElement,
log: (m) => t.diagnostic(m),
});
const { page, appRoot, teardown } = session;
const pickSelector = fixture.runtime.pickSelector || 'h1.hero-title';
try {
await waitForHandshake(page);
const cases = [
// Stamped with another app's root: dropped at load time, before
// any server roundtrip.
{ id: 'f0e1d2c3', appRoot: '/somewhere/else/entirely' },
// Legacy shape without a stamp: dropped when the server refuses
// its first checkpoint as unknown_session.
{ id: 'deadf00d' },
];
for (const foreign of cases) {
t.diagnostic(`Seeding foreign session ${foreign.id}${foreign.appRoot ? ' (stamped)' : ' (legacy shape)'}`);
await page.evaluate((saved) => {
localStorage.setItem('impeccable-live-session', JSON.stringify(saved));
localStorage.removeItem('impeccable-live-session-handled');
}, {
id: foreign.id,
state: 'CYCLING',
expected: 3,
arrived: 3,
visible: 1,
sourceFile: 'src/App.jsx',
pageUrl: '/',
checkpointRevision: 5,
...(foreign.appRoot ? { appRoot: foreign.appRoot } : {}),
});
await page.reload({ waitUntil: 'domcontentloaded' });
await waitForHandshake(page);
const deadline = Date.now() + 20_000;
for (;;) {
const current = await readLiveSessionStorage(page);
if (!current || current.id !== foreign.id) break;
if (Date.now() > deadline) throw new Error(`foreign session ${foreign.id} was never cleared`);
await new Promise((resolve) => setTimeout(resolve, 250));
}
const sessionsDir = join(appRoot, '.impeccable/live/sessions');
assert.equal(
existsSync(join(sessionsDir, `${foreign.id}.jsonl`)),
false,
`no ghost journal for ${foreign.id} may materialize in this project's store`,
);
}
// The surface is genuinely back: a fresh pick must work.
await pickElement(page, pickSelector, { position: fixture.runtime.pickPosition });
} finally {
await teardownAndResetBrowser(teardown);
}
});
}
if (shouldRunScenario('manual') && Array.isArray(fixture.runtime.manualEditScenarios) && fixture.runtime.manualEditScenarios.length > 0) {
const manualScenarioFilter = process.env.IMPECCABLE_E2E_MANUAL_SCENARIO || '';
for (const scenario of fixture.runtime.manualEditScenarios) {
if (manualScenarioFilter && !scenario.name.includes(manualScenarioFilter)) continue;
it(`Edit copy → Save → Apply/commit: ${scenario.name}`, liveE2eTestOptions, async (t) => {
const manualAgent = await createManualScenarioAgent(t, scenario);
if (!manualAgent) return;
const { agent, agentMode, probeState } = manualAgent;
const session = await bootFixtureSession({
name,
fixture,
browser,
agent,
wrapTarget: agentMode === 'llm' ? wrapTargetFromPickedElement : undefined,
log: (m) => t.diagnostic(m),
});
const { page, teardown } = session;
try {
await waitForHandshake(page);
if (fixture.runtime.preActions) await runPreActions(page, fixture.runtime.preActions);
const stages = Array.isArray(scenario.stages) ? scenario.stages : [scenario];
for (const stage of stages) {
await runManualEditStage(page, stage, {
t,
fixture,
session,
agentMode,
defaultSelector: stage.element?.selector || fixture.runtime.pickSelector || 'h1.hero-title',
});
}
if (scenario.probeMalformedAckBeforeApply) {
assert.equal(probeState?.malformedAckRejected, true, 'malformed manual Apply ack should fail loudly');
assert.equal(probeState?.applyCalls, 1, 'manual_edit_apply event should not be redelivered after the correct ack');
}
} finally {
await teardownAndResetBrowser(teardown);
}
});
}
}
if (shouldRunScenario('annotations') && fixture.runtime.liveChrome?.annotations) {
it('uploads annotations with generate and still accepts the variant', liveE2eTestOptions, async (t) => {
if (manualOnly || process.env.IMPECCABLE_E2E_MANUAL_SCENARIO) {
t.skip('manual scenario filter is active');
return;
}
const agentMode = process.env.IMPECCABLE_E2E_AGENT || 'fake';
const recordedGenerateEvents = [];
let baseAgent;
if (agentMode === 'llm') {
const llmConfig = resolveLlmAgentConfig({
model: process.env.IMPECCABLE_E2E_LLM_MODEL,
});
baseAgent = await createLlmAgent({
config: llmConfig,
log: (m) => t.diagnostic('[llm] ' + m),
});
if (!baseAgent) {
t.skip(`IMPECCABLE_E2E_AGENT=llm with provider=${llmConfig.provider} requires ${llmConfig.requiredEnv}`);
return;
}
t.diagnostic(`Using LLM agent (provider=${llmConfig.provider} model=${llmConfig.model})`);
} else {
baseAgent = createFakeAgent();
}
const agent = recordGenerateEvents(baseAgent, recordedGenerateEvents);
const session = await bootFixtureSession({
name,
fixture,
browser,
agent,
wrapTarget: wrapTargetFromPickedElement,
log: (m) => t.diagnostic(m),
});
const { page, teardown } = session;
const annotation = fixture.runtime.liveChrome.annotations;
const pickSelector = annotation.selector || fixture.runtime.pickSelector || 'h1.hero-title';
try {
await waitForHandshake(page);
if (fixture.runtime.preActions) await runPreActions(page, fixture.runtime.preActions);
await pickElement(page, pickSelector, { resetPickMode: true });
await drawAnnotationPinAndStroke(page, {
comment: annotation.comment || 'Make this selected element easier to scan',
});
await clickGo(page);
await waitForCyclingRobust(page, 3, {
agentMode,
preActions: fixture.runtime.preActions,
log: (m) => t.diagnostic(m),
});
const generateEvent = recordedGenerateEvents.at(-1);
await assertAnnotationUploadEvent(generateEvent);
assert.ok(existsSync(generateEvent.screenshotPath), 'annotation screenshot file exists');
assert.match(generateEvent.screenshotPath, /\.impeccable\/live\/annotations\//, 'annotation screenshot is stored under live annotations');
const sourceFile = await locateSessionFile(session.appRoot);
const svelteComponentTarget = svelteComponentTargetFor(sourceFile);
await clickNext(page);
assert.equal(await getVisibleVariant(page), 2, 'variant 2 visible after annotated generate');
await clickAccept(page, { expectedVariant: 2 });
await waitForBarHidden(page);
await waitForSourceClean(sourceFile, 20_000, { svelteComponentTarget });
} finally {
await teardownAndResetBrowser(teardown);
}
});
}
if (shouldRunScenario('exit') && fixture.runtime.liveChrome?.bottomBar) {
it('Exit removes live chrome cleanly', liveE2eTestOptions, async (t) => {
if (manualOnly || process.env.IMPECCABLE_E2E_MANUAL_SCENARIO) {
t.skip('manual scenario filter is active');
return;
}
const session = await bootFixtureSession({
name,
fixture,
browser,
agent: createFakeAgent(),
wrapTarget: wrapTargetFromPickedElement,
log: (m) => t.diagnostic(m),
});
try {
await waitForHandshake(session.page);
await clickExitLiveMode(session.page);
} finally {
await session.teardown();
}
});
}
});
}
// ---------------------------------------------------------------------------
// Helpers
// ---------------------------------------------------------------------------
/**
* The params and failure-injection scenarios assert on deterministic variant
* content (exact CSS values, exact font weights). An LLM agent legitimately
* produces neither, so they only run against the fake agent.
*/
function canRunFakeAgentScenario(t) {
if (manualOnly || process.env.IMPECCABLE_E2E_MANUAL_SCENARIO) {
t.skip('manual scenario filter is active');
return false;
}
if ((process.env.IMPECCABLE_E2E_AGENT || 'fake') !== 'fake') {
t.skip('scenario asserts on deterministic fake-agent output');
return false;
}
return true;
}
/**
* Boot a fixture straight to CYCLING on a component-preview session and hand
* back the handles the scenarios need: the manifest, the route source, and the
* session itself. Callers own teardown.
*/
async function bootComponentScenario({ t, name, fixture, expectedCount = 3, agent }) {
const session = await bootFixtureSession({
name,
fixture,
browser,
agent: agent || createFakeAgent(),
wrapTarget: wrapTargetFromPickedElement,
log: (m) => t.diagnostic(m),
});
try {
const { page, appRoot } = session;
await waitForHandshake(page);
if (fixture.runtime.preActions) await runPreActions(page, fixture.runtime.preActions);
await pickElement(page, fixture.runtime.pickSelector || 'h1.hero-title', {
position: fixture.runtime.pickPosition,
});
await clickGo(page);
await waitForCyclingRobust(page, expectedCount, {
agentMode: 'fake',
preActions: fixture.runtime.preActions,
log: (m) => t.diagnostic(m),
});
const manifestPath = await locateSessionFile(appRoot);
const svelteComponentTarget = svelteComponentTargetFor(manifestPath);
assert.ok(svelteComponentTarget, `expected a component-preview session, got ${manifestPath}`);
return {
session,
manifestPath,
svelteComponentTarget,
sourceFile: manifestPath,
};
} catch (err) {
await teardownAndResetBrowser(session.teardown);
throw err;
}
}
function readManifest(manifestPath) {
return JSON.parse(readFileSync(manifestPath, 'utf-8'));
}
/**
* The file the browser actually imports for a variant: the server snapshots
* every publish into a fresh `r<N>/` dir and points the manifest at it, so the
* session dir's copy is not what a mount reads.
*/
function revisionVariantPath(appRoot, manifestPath, variantNum) {
const manifest = readManifest(manifestPath);
const dir = manifest.revisionDir || manifest.componentDir;
assert.ok(dir, 'manifest carries a preview directory');
const extension = manifest.componentExtension || 'svelte';
return join(appRoot, dir, `v${variantNum}.${extension}`);
}
/** Poll the session journal for events of a given type. */
async function waitForJournalEvent(appRoot, sessionId, type, timeoutMs = 15_000) {
const journalPath = join(appRoot, '.impeccable', 'live', 'sessions', `${sessionId}.jsonl`);
const deadline = Date.now() + timeoutMs;
let seen = [];
while (Date.now() < deadline) {
seen = readJournalEvents(journalPath).filter((event) => event?.type === type);
if (seen.length > 0) return seen;
await new Promise((resolve) => setTimeout(resolve, 100));
}
throw new Error(
`no ${type} event in ${journalPath} after ${timeoutMs}ms; saw types: ${
[...new Set(readJournalEvents(journalPath).map((e) => e?.type))].join(', ') || '(none)'
}`,
);
}
function readJournalEvents(journalPath) {
if (!existsSync(journalPath)) return [];
return readFileSync(journalPath, 'utf-8')
.split('\n')
.filter(Boolean)
.map((line) => {
try { return JSON.parse(line)?.event || null; } catch { return null; }
})
.filter(Boolean);
}
/**
* Assert the roots live.mjs resolved for an `appDir` fixture.
*
* The boot ran from the staged repo root, which carries no dev-server config
* of its own. Live mode has to pick the nested app, keep PRODUCT.md/DESIGN.md
* discovery at the git root, and leave its session state under the app plus a
* pointer at the repo root. A repo-root server.json means the session forked
* into a second, empty project — the failure this fixture exists to catch.
*/
function assertNestedAppRoots(session) {
const { tmp, appRoot, appDir, liveBoot } = session;
assert.ok(liveBoot, 'appDir fixtures boot through live.mjs and keep its payload');
assert.equal(liveBoot.ok, true, `live.mjs boot payload: ${JSON.stringify(liveBoot)}`);
const roots = liveBoot.roots || {};
assert.ok(
roots.appRoot && roots.appRoot.endsWith(`/${appDir}`),
`roots.appRoot should end with /${appDir}, got ${roots.appRoot}`,
);
assertSamePath(roots.appRoot, appRoot, 'roots.appRoot');
assertSamePath(roots.repoRoot, tmp, 'roots.repoRoot');
assertSamePath(roots.contextRoot, tmp, 'roots.contextRoot');
assertSamePath(liveBoot.projectRoot, appRoot, 'projectRoot');
assert.equal(liveBoot.hasProduct, true, 'PRODUCT.md at the repo root is read from the nested app');
assert.equal(liveBoot.hasDesign, true, 'DESIGN.md at the repo root is read from the nested app');
assert.ok(
existsSync(join(appRoot, '.impeccable/live/roots.json')),
'the resolved manifest is persisted under the app',
);
assert.ok(
existsSync(join(appRoot, '.impeccable/live/server.json')),
'the live server registers itself under the app',
);
assert.ok(
existsSync(join(tmp, '.impeccable/live/app-root.json')),
'the repo root gets a pointer to the app',
);
assert.equal(
existsSync(join(tmp, '.impeccable/live/server.json')),
false,
'no live session state may be written at the repo root',
);
}
function assertSamePath(actual, expected, label) {
const resolve = (p) => {
try { return realpathSync(String(p)); } catch { return String(p); }
};
assert.equal(resolve(actual), resolve(expected), `${label}: ${actual} !== ${expected}`);
}
function recordGenerateEvents(agent, events) {
return {
...agent,
async generateVariants(event, context) {
events.push(event);
return agent.generateVariants(event, context);
},
};
}
function countSourceVariants(source) {
return (String(source).match(/<div\s+data-impeccable-variant="(?!original")/g) || []).length;
}
function escapeRegExp(value) {
return String(value).replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
}
async function waitForGenerationTimings(tmp, id, { timeoutMs = 5_000, requireAllVariants = true } = {}) {
const snapshotPath = join(tmp, '.impeccable', 'live', 'sessions', `${id}.snapshot.json`);
const journalPath = join(tmp, '.impeccable', 'live', 'sessions', `${id}.jsonl`);
const deadline = Date.now() + timeoutMs;
let lastTimings = null;
while (Date.now() < deadline) {
if (existsSync(snapshotPath)) {
const snapshot = JSON.parse(readFileSync(snapshotPath, 'utf-8'));
const timings = snapshot.generationTimings || {};
lastTimings = timings;
if (timings.generation_ready && timings.first_reviewable && (!requireAllVariants || timings.all_variants_ready)) return timings;
}
await new Promise((resolve) => setTimeout(resolve, 50));
}
const checkpointReasons = existsSync(journalPath)
? readFileSync(journalPath, 'utf-8')
.split('\n')
.filter(Boolean)
.map((line) => JSON.parse(line)?.event)
.filter((event) => event?.type === 'checkpoint')
.map((event) => ({ reason: event.reason, arrivedVariants: event.arrivedVariants, expectedVariants: event.expectedVariants }))
: [];
throw new Error(`generation timings did not complete for ${id}: timings=${JSON.stringify(lastTimings)} checkpoints=${JSON.stringify(checkpointReasons)}`);
}
async function captureLiveE2eFailure({ name, fixture, session, sourceFile, error, log = () => {} }) {
const root = process.env.IMPECCABLE_E2E_ARTIFACT_DIR;
if (!root || !session?.tmp) return;
try {
const tmp = session.tmp;
const dir = join(root, `${safeArtifactName(name)}-${Date.now()}`);
mkdirSync(dir, { recursive: true });
writeFileSync(join(dir, 'error.txt'), String(error?.stack || error?.message || error || ''), 'utf-8');
writeFileSync(join(dir, 'fixture.json'), JSON.stringify(fixture, null, 2), 'utf-8');
writeFileSync(join(dir, 'console-errors.log'), (session.consoleErrors || []).join('\n'), 'utf-8');
writeFileSync(
join(dir, 'failed-requests.log'),
(session.failedRequests || []).map((f) => `${f.reason} ${f.url}`).join('\n'),
'utf-8',
);
writeFileSync(join(dir, 'dev-server.log'), session.dev?.log?.() || '', 'utf-8');
writeCommandOutput(dir, 'git-status.txt', tmp, ['status', '--short']);
writeCommandOutput(dir, 'git-diff.patch', tmp, ['diff', '--', '.']);
const locatedSource = sourceFile || await locateSessionFile(tmp).catch(() => null);
if (locatedSource && existsSync(locatedSource)) {
writeFileSync(join(dir, 'source-file.txt'), relative(tmp, locatedSource), 'utf-8');
copyFileFromTmp(tmp, locatedSource, join(dir, 'sources'));
const sourceShadow = sourceShadowTargetFor(locatedSource);
if (sourceShadow && existsSync(sourceShadow)) copyFileFromTmp(tmp, sourceShadow, join(dir, 'sources'));
const svelteTarget = svelteComponentTargetFor(locatedSource);
if (svelteTarget?.sourceFile && existsSync(svelteTarget.sourceFile)) {
copyFileFromTmp(tmp, svelteTarget.sourceFile, join(dir, 'sources'));
}
}
for (const file of walkSources(tmp)) copyFileFromTmp(tmp, file, join(dir, 'sources'));
copyDirIfExists(join(tmp, '.impeccable', 'live'), join(dir, 'impeccable-live'));
copyDirIfExists(join(tmp, 'node_modules', '.impeccable-live'), join(dir, 'impeccable-live-preview'));
copyDirIfExists(join(tmp, '.impeccable', 'live', 'previews'), join(dir, 'impeccable-live-previews'));
if (session.page) {
const html = await withCaptureTimeout(session.page.content(), 5_000, 'page content').catch((err) => `capture failed: ${err.message}`);
writeFileSync(join(dir, 'page.html'), html, 'utf-8');
await withCaptureTimeout(
session.page.screenshot({ path: join(dir, 'page.png'), fullPage: true }),
5_000,
'page screenshot',
).catch((err) => writeFileSync(join(dir, 'screenshot-error.txt'), err.message, 'utf-8'));
}
log(`Failure artifacts written to ${dir}`);
} catch (captureErr) {
log(`Failure artifact capture failed: ${captureErr.message}`);
}
}
function writeCommandOutput(dir, fileName, cwd, args) {
try {
const output = execFileSync('git', args, { cwd, encoding: 'utf-8' });
writeFileSync(join(dir, fileName), output, 'utf-8');
} catch (err) {
writeFileSync(join(dir, fileName), [err.stdout, err.stderr, err.message].filter(Boolean).join('\n'), 'utf-8');
}
}
function copyFileFromTmp(tmp, file, destRoot) {
const rel = relative(tmp, file);
if (!rel || rel.startsWith('..')) return;
const dest = join(destRoot, rel);
mkdirSync(dirname(dest), { recursive: true });
cpSync(file, dest);
}
function copyDirIfExists(from, to) {
if (!existsSync(from)) return;
mkdirSync(dirname(to), { recursive: true });
cpSync(from, to, { recursive: true });
}
function safeArtifactName(name) {
return String(name || 'fixture').replace(/[^a-z0-9._-]+/gi, '-').replace(/^-+|-+$/g, '') || 'fixture';
}
function withCaptureTimeout(promise, timeoutMs, label) {
let timer;
const timeout = new Promise((_, reject) => {
timer = setTimeout(() => reject(new Error(`${label} timed out after ${timeoutMs}ms`)), timeoutMs);
});
return Promise.race([promise, timeout]).finally(() => clearTimeout(timer));
}
async function createManualScenarioAgent(t, scenario = {}) {
const requested = (process.env.IMPECCABLE_E2E_MANUAL_AGENT || process.env.IMPECCABLE_E2E_AGENT || 'auto')
.trim()
.toLowerCase();
if (requested === 'fake' || requested === 'mock') {
t.diagnostic('Using fake agent for manual-edit scenarios (explicit fallback)');
const probeState = {};
return {
agent: maybeWrapMalformedAckProbe(createFakeAgent(), scenario, probeState, t),
agentMode: 'fake',
probeState,
};
}
if (requested !== 'auto' && requested !== 'llm') {
throw new Error(`Unsupported manual-edit e2e agent: ${requested}`);
}
const llmConfig = resolveLlmAgentConfig({
model: process.env.IMPECCABLE_E2E_LLM_MODEL,
});
const agent = await createLlmAgent({
config: llmConfig,
log: (m) => t.diagnostic('[llm] ' + m),
});
if (agent) {
t.diagnostic(`Using LLM agent for manual-edit scenarios (provider=${llmConfig.provider} model=${llmConfig.model})`);
const probeState = {};
return {
agent: maybeWrapMalformedAckProbe(agent, scenario, probeState, t),
agentMode: 'llm',
probeState,
};
}
if (requested === 'llm') {
t.skip(`IMPECCABLE_E2E_AGENT=llm with provider=${llmConfig.provider} requires ${llmConfig.requiredEnv}`);
return null;
}
t.diagnostic(`Using fake agent for manual-edit scenarios because ${llmConfig.requiredEnv} is unset`);
const probeState = {};
return {
agent: maybeWrapMalformedAckProbe(createFakeAgent(), scenario, probeState, t),
agentMode: 'fake',
probeState,
};
}
function maybeWrapMalformedAckProbe(agent, scenario, probeState, t) {
if (!scenario.probeMalformedAckBeforeApply) return agent;
return {
...agent,
async applyManualEdits(event, context = {}) {
probeState.applyCalls = (probeState.applyCalls || 0) + 1;
const sourceFile = firstExpectedSourceFile(scenario) || 'src/App.jsx';
try {
runEngineSync('live-poll', ['--reply', 'done', '--file', sourceFile], {
cwd: context.tmp,
stdio: ['ignore', 'pipe', 'pipe'],
});
assert.fail('malformed manual Apply ack unexpectedly succeeded');
} catch (err) {
const output = [err.stdout, err.stderr, err.message].filter(Boolean).join('\n');
assert.match(output, /--reply EVENT_ID done|must be the event id|Missing reply status/);
probeState.malformedAckRejected = true;
}
const buffer = JSON.parse(readFileSync(join(context.tmp, '.impeccable/live/pending-manual-edits.json'), 'utf-8'));
assert.ok(buffer.entries.length > 0, 'malformed ack must not clear staged manual edits');
t.diagnostic(`Malformed manual Apply ack rejected for ${event.id}; continuing with correct reply`);
return agent.applyManualEdits(event, context);
},
};
}
function firstExpectedSourceFile(scenario) {
const stages = Array.isArray(scenario.stages) ? scenario.stages : [scenario];
for (const stage of stages) {
for (const edit of stage.edits || []) {
if (edit.expectedSourceFile) return edit.expectedSourceFile;
}
}
return null;
}
function wrapTargetFromPickedElement(event) {
const element = event.element || {};
const tag = typeof element.tagName === 'string'
? element.tagName.trim().toLowerCase()
: '';
const classes = Array.isArray(element.classes)
? element.classes.filter(Boolean).join(' ')
: typeof element.className === 'string'
? element.className.trim().split(/\s+/).filter(Boolean).join(' ')
: extractClassAttr(element.outerHTML);
const elementId = typeof element.id === 'string' ? element.id.trim() : '';
return {
tag: tag || 'h1',
...(classes ? { classes } : {}),
...(elementId ? { elementId } : {}),
...(element.textContent ? { text: String(element.textContent).trim() } : {}),
};
}
function extractClassAttr(outerHTML) {
if (typeof outerHTML !== 'string') return '';
const match = outerHTML.match(/\sclass=(["'])(.*?)\1/);
return match ? match[2].trim().split(/\s+/).filter(Boolean).join(' ') : '';
}
async function runManualScenarioActions(page, actions, { t, fixture, session, defaultSelector, agentMode }) {
for (const action of actions || []) {
if (action.type === 'variantAccept') {
await runAcceptedVariantCycle(page, {
t,
fixture,
session,
pickSelector: action.selector || defaultSelector,
pickFirst: true,
agentMode,
});
continue;
}
if (action.type === 'acceptCurrentSelection') {
await runAcceptedVariantCycle(page, {
t,
fixture,
session,
pickSelector: defaultSelector,
pickFirst: false,
agentMode,
});
continue;
}
await runPreActions(page, [action]);
}
}
async function runManualEditStage(page, stage, { t, fixture, session, agentMode, defaultSelector }) {
const tmp = session.appRoot;
if (stage.beforeManualEdit) {
await runManualScenarioActions(page, stage.beforeManualEdit, {
t,
fixture,
session,
defaultSelector,
agentMode,
});
}
await pickElement(
page,
stage.element?.selector || defaultSelector,
{ position: stage.element?.position, resetPickMode: true },
);
t.diagnostic('Manual scenario clicking Edit copy');
await clickEditCopy(page);
for (const edit of stage.edits || []) {
await editTextLeaf(page, edit.leafSelector, edit.newText);
}
t.diagnostic('Manual scenario clicking Save');
await clickSaveEdit(page);
const expectedStashCount = stage.expectedStashCount || Math.max(1, stage.edits?.length || 1);
await assertApplyDockVisible(page, expectedStashCount, {
timeout: agentMode === 'llm' ? 20_000 : 5_000,
});
assert.equal(
await getServerManualEditStashCount(session.live),
expectedStashCount,
'manual edit stash count after Save',
);
if (stage.afterSave) {
await runManualScenarioActions(page, stage.afterSave, {
t,
fixture,
session,
defaultSelector,
agentMode,
});
}
if (stage.skipApply === true) {
assert.equal(
await getServerManualEditStashCount(session.live),
stage.expectedFinalStashCount ?? 0,
'manual edit stash count after scenario action',
);
return;
}
t.diagnostic('Manual scenario clicking Apply/commit');
await clickApplyEdits(page);
if (stage.expectApplyLoading) {
await assertApplyDockLoading(page, {
timeout: agentMode === 'llm' ? 20_000 : 5_000,
});
}
const applyTimeoutMs = stage.applyTimeoutMs || (agentMode === 'llm' ? 120_000 : 20_000);
await waitForServerManualEditStashCount(session.live, 0, {
timeout: applyTimeoutMs,
});
await waitForApplyDockHidden(page, { timeout: 10_000 });
const remaining = await getServerManualEditStashCount(session.live);
assert.equal(remaining, 0, 'manual edit stash cleared after Apply');
for (const edit of stage.edits || []) {
if (edit.expectedVisibleText) {
try {
await assertVisibleText(page, edit.leafSelector, edit.expectedVisibleText, {
timeout: agentMode === 'llm' ? 60_000 : 20_000,
});
} catch (err) {
if (edit.expectedSourceFile) {
t.diagnostic(`--- source ${edit.expectedSourceFile} after visible-text failure ---`);
t.diagnostic(readFileSync(join(tmp, edit.expectedSourceFile), 'utf-8'));
}
throw err;
}
}
}
for (const edit of stage.edits || []) {
if (edit.expectedSourceFile) {
assertSourceApplied(
tmp,
edit.expectedSourceFile,
edit.expectOriginalRemaining ? '' : edit.originalText,
edit.expectedSourceMatch || edit.newText,
);
for (const snippet of edit.expectedSourceAlso || []) {
assertSourceContains(tmp, edit.expectedSourceFile, snippet);
}
for (const pattern of edit.expectedSourceRegex || []) {
assertSourceMatches(tmp, edit.expectedSourceFile, pattern);
}
for (const snippet of edit.expectedSourceMissing || []) {
assertSourceMissing(tmp, edit.expectedSourceFile, snippet);
}
}
}
if (stage.expectNoRollback) {
const status = await getServerManualEditStatus(session.live);
const rolledBackFiles = status.manualEdits?.lastActivity?.rolledBackFiles || [];
assert.deepEqual(rolledBackFiles, [], 'manual Apply should not report rolled-back files');
assert.notEqual(status.manualEdits?.lastActivity?.reason, 'manual_edit_repair_needs_decision');
}
if (stage.refreshAfterApply) {
await page.reload({ waitUntil: 'domcontentloaded' });
await waitForHandshake(page);
for (const edit of stage.edits || []) {
if (edit.expectedVisibleText) {
await assertVisibleText(page, edit.leafSelector, edit.expectedVisibleText, {
timeout: agentMode === 'llm' ? 60_000 : 20_000,
});
}
}
}
if (stage.afterApply) {
await runManualScenarioActions(page, stage.afterApply, {
t,
fixture,
session,
defaultSelector,
agentMode,
});
}
}
async function runAcceptedVariantCycle(page, { t, fixture, session, pickSelector, pickFirst, agentMode }) {
if (pickFirst) {
t.diagnostic(`Manual scenario picking ${pickSelector} before variant accept`);
await pickElement(page, pickSelector, { resetPickMode: true });
}
t.diagnostic('Manual scenario clicking Go');
await clickGo(page);
await waitForCycling(page, 3, {
timeout: agentMode === 'llm' ? 240_000 : 30_000,
});
await clickNext(page);
assert.equal(await getVisibleVariant(page), 2, 'variant 2 visible before manual scenario accept');
await clickAccept(page, { expectedVariant: 2 });
const sourceFile = await locateSessionFile(session.appRoot);
await waitForSourceClean(sourceFile, 20_000);
await waitForBarHidden(page, { timeout: 10_000 }).catch(() => {});
const expectSelector = fixture.runtime.reloadProbe?.expectSelector || pickSelector;
await waitForAcceptedSelectionReady(page, expectSelector, {
timeout: agentMode === 'llm' ? 60_000 : 20_000,
});
}
async function waitForAcceptedSelectionReady(page, selector, { timeout }) {
await page.waitForFunction(
(sel) => {
const all = document.querySelectorAll(sel);
if (all.length < 1) return false;
for (const el of all) {
if (el.closest('[data-impeccable-variants],[data-impeccable-variant]')) return false;
}
return true;
},
selector,
{ timeout },
);
}
async function readLiveSessionStorage(page) {
return page.evaluate(() => {
const raw = localStorage.getItem('impeccable-live-session');
return raw ? JSON.parse(raw) : null;
});
}
async function waitForVariantCounter(page, variant, count, { timeout = 15_000 } = {}) {
try {
await page.waitForFunction(
({ variant, count }) => {
const query = window.__impeccableLiveQuery || ((sel) => document.querySelector(sel));
const bar = query('#impeccable-live-bar');
const text = bar?.textContent || '';
return text.includes(`${variant}/${count}`);
},
{ variant, count },
{ timeout },
);
} catch (err) {
const snapshot = await page.evaluate(() => {
const query = window.__impeccableLiveQuery || ((sel) => document.querySelector(sel));
const bar = query('#impeccable-live-bar');
const wrapper = document.querySelector('[data-impeccable-variants]');
return {
barText: bar?.textContent || null,
debugState: window.__IMPECCABLE_LIVE_CHROME_CORE__?.debugState?.() || null,
storage: localStorage.getItem('impeccable-live-session'),
wrapper: wrapper ? { preview: wrapper.dataset.impeccablePreview, count: wrapper.dataset.impeccableVariantCount, html: wrapper.outerHTML.slice(0, 500) } : null,
};
}).catch((snapErr) => ({ error: snapErr.message }));
console.error('--- waitForVariantCounter snapshot ---\n' + JSON.stringify(snapshot, null, 2));
err.message += '\nVariant counter snapshot: ' + JSON.stringify(snapshot, null, 2);
throw err;
}
}
async function waitForRecoverableVariantSession(page, variant, count, { timeout = 15_000 } = {}) {
await page.waitForFunction(
({ variant, count }) => {
const query = window.__impeccableLiveQuery || ((sel) => document.querySelector(sel));
const bar = query('#impeccable-live-bar');
const text = bar?.textContent || '';
const raw = localStorage.getItem('impeccable-live-session');
let saved = null;
try { saved = raw ? JSON.parse(raw) : null; } catch {}
return Boolean(
saved
&& saved.visible === variant
&& saved.expected === count
&& saved.previewMode === 'svelte-component'
&& /Reveal the selected element to resume/i.test(text)
);
},
{ variant, count },
{ timeout },
);
}
async function waitForAcceptedDom(page, selector, { allowVariantRoot = false, timeout = 20_000 } = {}) {
await page.waitForFunction(
({ sel, allowVariantRoot }) => {
const all = document.querySelectorAll(sel);
if (all.length < 1) return false;
for (const el of all) {
if (el.closest('[data-impeccable-variants]')) return false;
if (el.closest('[data-impeccable-carbonize]')) return false;
if (!allowVariantRoot && el.closest('[data-impeccable-variant]')) return false;
}
return true;
},
{ sel: selector, allowVariantRoot },
{ timeout },
);
}
function assertSourceMissing(tmp, file, text) {
const full = join(tmp, file);
const body = readFileSync(full, 'utf-8');
assert.equal(
body.includes(text),
false,
`source ${file} should not contain discarded text ${JSON.stringify(text)}`,
);
}
function assertSourceContains(tmp, file, text) {
const full = join(tmp, file);
const body = readFileSync(full, 'utf-8');
assert.equal(
body.includes(text),
true,
`source ${file} should still contain ${JSON.stringify(text)}`,
);
}
function assertSourceMatches(tmp, file, pattern) {
const full = join(tmp, file);
const body = readFileSync(full, 'utf-8');
const re = new RegExp(pattern);
assert.equal(
re.test(body),
true,
`source ${file} should match ${pattern}`,
);
}
async function assertVisibleText(page, selector, text, { timeout = 20_000 } = {}) {
try {
await page.waitForFunction(
({ sel, expected }) => {
const el = document.querySelector(sel);
return Boolean(el && (el.textContent || '').includes(expected));
},
{ sel: selector, expected: text },
{ timeout },
);
} catch (err) {
const actual = await page.evaluate((sel) => document.querySelector(sel)?.textContent || null, selector).catch(() => null);
throw new Error(`visible text ${selector} did not include ${JSON.stringify(text)}; actual=${JSON.stringify(actual)}; ${err.message}`);
}
}
async function readVisibleVariantForCycle(page, { timeout = 5_000 } = {}) {
const start = Date.now();
let last = null;
while (Date.now() - start < timeout) {
last = await getVisibleVariant(page);
if (Number.isInteger(last) && last > 0) return last;
await page.waitForTimeout(250);
}
return last;
}
async function evaluatePageWithTimeout(page, fn, arg, timeoutMs, label) {
let timer;
const timeout = new Promise((_, reject) => {
timer = setTimeout(() => reject(new Error(`${label} timed out after ${timeoutMs}ms`)), timeoutMs);
});
return Promise.race([page.evaluate(fn, arg), timeout]).finally(() => clearTimeout(timer));
}
async function getServerManualEditStashCount(live, pageUrl = '/') {
const res = await fetch(
`http://localhost:${live.port}/manual-edit-stash?token=${encodeURIComponent(live.token)}&pageUrl=${encodeURIComponent(pageUrl)}`,
);
if (!res.ok) throw new Error(`manual-edit-stash count failed: ${res.status}`);
const body = await res.json();
return body.count || 0;
}
async function getServerManualEditStatus(live) {
const res = await fetch(`http://localhost:${live.port}/status?token=${encodeURIComponent(live.token)}`);
if (!res.ok) throw new Error(`manual edit status failed: ${res.status}`);
return res.json();
}
async function waitForServerManualEditStashCount(live, expectedCount, { pageUrl = '/', timeout = 20_000 } = {}) {
const start = Date.now();
let last = null;
let lastError = null;
let lastActivity = null;
let lastStatusCheck = 0;
while (Date.now() - start < timeout) {
try {
last = await getServerManualEditStashCount(live, pageUrl);
lastError = null;
if (last === expectedCount) return;
if (expectedCount === 0 && Date.now() - lastStatusCheck > 1_000) {
lastStatusCheck = Date.now();
lastActivity = (await getServerManualEditStatus(live)).manualEdits?.lastActivity || null;
if (lastActivity?.type === 'manual_edit_repair_needs_decision') {
throw new Error(`manual edit Apply needs repair decision before stash cleared; last=${last}; lastActivity=${JSON.stringify(lastActivity)}`);
}
}
} catch (err) {
lastError = err;
if (/manual edit Apply needs repair decision/.test(err.message || '')) throw err;
}
await new Promise((resolve) => setTimeout(resolve, 250));
}
try {
lastActivity = (await getServerManualEditStatus(live)).manualEdits?.lastActivity || null;
} catch {}
throw new Error(`manual edit stash count did not reach ${expectedCount}; last=${last}; lastError=${lastError?.message || 'none'}; lastActivity=${JSON.stringify(lastActivity)}`);
}
async function clickPickToggle(page, selector) {
try {
await page.locator(selector).click({ timeout: 5_000 });
return;
} catch (err) {
const clicked = await page.evaluate((sel) => {
const btn = document.querySelector(sel);
if (!btn) return false;
btn.click();
return true;
}, selector);
if (!clicked) throw err;
}
}
/**
* Poll the file until carbonize cleanup has landed: no variants wrapper, no
* carbonize markers, no leftover variant divs. Returns the final contents.
*/
async function waitForSourceClean(filePath, timeoutMs, { svelteComponentTarget: knownSvelteTarget = null } = {}) {
const start = Date.now();
let last = '';
const shadowTarget = sourceShadowTargetFor(filePath);
const svelteTarget = knownSvelteTarget || svelteComponentTargetFor(filePath);
if (shadowTarget) {
let handled = false;
while (Date.now() - start < timeoutMs) {
last = readFileSync(filePath, 'utf-8');
if (last.includes('source-shadow preview handled')) {
handled = true;
break;
}
await new Promise((r) => setTimeout(r, 100));
}
if (!handled) {
throw new Error(`source-shadow preview not handled after ${timeoutMs}ms — last contents:\n${last}`);
}
filePath = shadowTarget;
} else if (svelteTarget) {
filePath = svelteTarget.sourceFile;
}
while (Date.now() - start < timeoutMs) {
last = readFileSync(filePath, 'utf-8');
const dirty =
(svelteTarget && existsSync(svelteTarget.manifestPath)) ||
last.includes('data-impeccable-variants=') ||
last.includes('impeccable-variants-start') ||
last.includes('impeccable-carbonize-start') ||
last.includes('data-impeccable-carbonize=') ||
last.includes('data-impeccable-variant=');
if (!dirty) return last;
await new Promise((r) => setTimeout(r, 100));
}
throw new Error(`source not clean after ${timeoutMs}ms — last contents:\n${last}`);
}
function sourceShadowTargetFor(filePath) {
let body;
try { body = readFileSync(filePath, 'utf-8'); } catch { return null; }
if (!body.includes('data-impeccable-preview="source-shadow"')) return null;
const match = body.match(/\bdata-impeccable-source-file=(["'])(.*?)\1/);
if (!match) return null;
const root = filePath.includes('/.impeccable/')
? filePath.slice(0, filePath.indexOf('/.impeccable/'))
: dirname(filePath);
return join(root, decodeHtmlAttr(match[2]));
}
function svelteComponentTargetFor(filePath) {
if (!filePath.endsWith('/manifest.json') && !filePath.endsWith('\\manifest.json')) return null;
let manifest;
try { manifest = JSON.parse(readFileSync(filePath, 'utf-8')); } catch { return null; }
if (manifest.previewMode !== 'svelte-component' || !manifest.sourceFile || !manifest.componentDir) return null;
const sep = pathSepFor(filePath);
const markers = [
`${sep}.impeccable${sep}live${sep}previews${sep}`,
`${sep}node_modules${sep}.impeccable-live${sep}`,
`${sep}src${sep}lib${sep}impeccable${sep}`,
`${sep}app${sep}.impeccable-live${sep}`,
];
const marker = markers.find((candidate) => filePath.includes(candidate));
const idx = marker ? filePath.indexOf(marker) : -1;
const root = idx === -1 ? dirname(dirname(dirname(dirname(dirname(filePath))))) : filePath.slice(0, idx);
return {
manifest,
manifestPath: filePath,
sourceFile: join(root, manifest.sourceFile),
expectedTag: expectedTagFromOriginalMarkup(manifest.originalMarkup),
};
}
function pathSepFor(filePath) {
return filePath.includes('\\') ? '\\' : '/';
}
function expectedTagFromOriginalMarkup(markup) {
const match = String(markup || '').match(/<([A-Za-z][\w:-]*)\b/);
return match ? match[1] : '[A-Za-z][\\w:-]*';
}
function fixtureUsesSvelteKitAdapter(fixture) {
return Array.isArray(fixture?.config?.files)
&& fixture.config.files.includes('src/app.html')
&& Array.isArray(fixture?.sourceFiles)
&& fixture.sourceFiles.some((file) => file.endsWith('.svelte'));
}
function decodeHtmlAttr(value) {
return String(value || '')
.replace(/&quot;/g, '"')
.replace(/&lt;/g, '<')
.replace(/&gt;/g, '>')
.replace(/&amp;/g, '&');
}
async function assertStateProbe(page, probe, label, { baseline = null } = {}) {
const snapshot = {};
if (probe.textSelector) {
const actual = await page.locator(probe.textSelector).first().textContent({ timeout: 5_000 });
snapshot.text = normalizeText(actual);
assert.equal(
snapshot.text,
normalizeText(probe.expectedText),
`stateProbe text ${label}`,
);
}
if (probe.windowProperty) {
const actual = await page.evaluate((prop) => window[prop], probe.windowProperty);
snapshot.windowValue = actual;
if (Object.hasOwn(probe, 'expectedWindowValue')) {
assert.equal(
actual,
probe.expectedWindowValue,
`stateProbe ${probe.windowProperty} ${label}`,
);
}
if (probe.expectWindowUnchanged && baseline) {
assert.equal(
actual,
baseline.windowValue,
`stateProbe ${probe.windowProperty} unchanged ${label}`,
);
}
}
return snapshot;
}
function normalizeText(value) {
return String(value || '').replace(/\s+/g, ' ').trim();
}
/**
* Find the source file that received the wrapper. We look for any tracked
* file containing the variants marker — the agent always writes to exactly
* one file per session.
*/
async function locateSessionFile(tmp) {
const candidates = walkSources(tmp);
for (const f of candidates) {
const body = readFileSync(f, 'utf-8');
if (
body.includes('data-impeccable-variants=') ||
body.includes('impeccable-carbonize-start') ||
body.includes('impeccable-variants-start')
) {
return f;
}
}
for (const f of walkComponentManifests(tmp)) {
const body = readFileSync(f, 'utf-8');
if (/"previewMode": "(?:svelte|vue)-component"/.test(body)) return f;
}
throw new Error('Could not locate session source file under ' + tmp);
}
function walkComponentManifests(root) {
const results = [];
const stack = [
join(root, '.impeccable/live/previews'),
join(root, 'node_modules/.impeccable-live'),
join(root, 'src/lib/impeccable'),
join(root, 'app/.impeccable-live'),
];
while (stack.length) {
const dir = stack.pop();
let entries;
try { entries = readdirSync(dir, { withFileTypes: true }); } catch { continue; }
for (const e of entries) {
const full = join(dir, e.name);
if (e.isDirectory()) {
stack.push(full);
} else if (e.name === 'manifest.json') {
results.push(full);
}
}
}
return results;
}
function walkSources(root) {
const results = [];
const stack = [root];
const SKIP = new Set(['node_modules', '.git', '.svelte-kit', 'dist', '.vite', 'build', '.next']);
const EXTS = ['.html', '.jsx', '.tsx', '.svelte', '.astro', '.vue'];
while (stack.length) {
const dir = stack.pop();
let entries;
try { entries = readdirSync(dir, { withFileTypes: true }); } catch { continue; }
for (const e of entries) {
const full = join(dir, e.name);
if (e.isDirectory()) {
if (!SKIP.has(e.name)) stack.push(full);
continue;
}
if (EXTS.some((x) => e.name.endsWith(x))) results.push(full);
}
}
return results;
}