Refactor: cleaner top-level directory structure (#138)

* refactor(content): merge content/site/ into site/content/

Phase 1 step 1 of the directory restructure. The dual content tree was
called out in CLAUDE.md as cleanup; both trees were already in sync
except for anti-patterns-catalog.js, which moves to site/data/.

- Delete content/site/skills/ and content/site/tutorials/ (duplicates of
  site/content/, which is what Astro's content collection actually reads).
- Move content/site/anti-patterns-catalog.js -> site/data/.
- Update scripts/lib/sub-pages-data.js and scripts/build.js to read from
  site/content/ and site/data/.
- Drop content/site/ from validateProse target list (site/content was
  already there).
- Rewrite the "Two content trees" section in CLAUDE.md as a single-tree
  pointer; update stale dev-server text mentioning the deleted
  server/index.js.

Tests: 186/186 pass. Skills build: clean.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(skill): rename source/skills/impeccable/ -> skill/

Phase 1 step 2 of the directory restructure. The path was redundantly
nested ("source/" wrapper plus "skills/impeccable/" — singular content
hidden behind the plural). Collapses to flat skill/SKILL.md +
skill/reference/ + skill/scripts/.

- Move source/skills/impeccable/ -> skill/.
- Rewrite scripts/lib/utils.js readSourceFiles(): drop the multi-skill
  iteration (CLAUDE.md commits to a single user-invocable skill); read
  skill/SKILL.md directly.
- Update scripts/build.js, scripts/generate-og-image.js, and the
  sub-pages data layer to point at skill/.
- Update tests/lib/utils.test.js: drop the "multi-skill" and "dir-name
  fallback" cases, update single-skill paths to skill/.
- Update tests/build.test.js similarly: drop "multiple skills"
  integration test, update paths.
- Update non-glob path joins in tests/framework-fixtures.test.mjs,
  tests/live-e2e/session.mjs, tests/live-e2e/agents/llm-agent.mjs,
  tools/live-loop.mjs.
- Update prose/text references in CLAUDE.md, AGENTS.md, DEVELOP.md,
  README.md, scripts/lib/sub-pages-data.js, bin/commands/skills.mjs,
  site/data/anti-patterns-catalog.js, site/pages/docs/[...slug].astro,
  docs/adr-live-variant-mode.md, docs/plans/.

Eval framework note: the separate impeccable-evals repo reads
../impeccable/source/skills/impeccable/ and needs a coordinated
rename to ../impeccable/skill/.

Tests: 186/186 pass. Skills build: clean.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor: rename docs/ -> notes/

Phase 1 step 3 of the directory restructure. The internal docs/ dir
(ADRs and plans) clashed with the site's /docs route. Renaming it
"notes/" makes the difference unambiguous: notes/ is project-internal
process, /docs is the user-facing route under site/pages/docs/.

No code references the dir; the rename is a clean git mv.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(site): move public/ under site/public/

Phase 2 step 4 of the directory restructure. Public assets and the
Astro publicDir now live alongside the rest of the site, so site/
is fully self-contained for static content.

- git mv public site/public.
- astro.config.mjs: add publicDir: './site/public'. Astro defaults to
  ./public at the project root, so the override is required.
- scripts/build.js: write generated _data, _headers, _redirects,
  _routes.json, and js/detect-antipatterns-browser.js into
  site/public/. Also delete the dead _REMOVED() Bun static-site
  builder (replaced by Astro at #130; the placeholder no longer earns
  its keep).
- scripts/build.js validateProse: replace the stale public/index.html
  reference (deleted at the Astro migration) with site/pages/index.astro
  in the count-validation file list, restoring homepage drift detection.
- scripts/generate-og-image.js: write OG image into site/public/.
- scripts/screenshot-antipatterns.js: read examples from + write
  screenshots to site/public/antipattern-{examples,images}/.
- scripts/lib/sub-pages-data.js: load command demos from
  site/public/js/demos/commands.
- .gitignore: rename the public/* generator-output entries to
  site/public/*.
- CLAUDE.md: refresh CSS/data-file paths (still pointing at the old
  pre-Astro public/css/ + public/js/ tree), point the changelog and
  command-add checklists at site/pages/index.astro and
  site/scripts/data.js + site/scripts/components/framework-viz.js.

Cloudflare Pages note: functions/ stays at the repo root because
CF Pages auto-discovers it there with no configuration knob to
relocate. Moving it under site/ would either break deployment or
require a build-time copy step that adds more complexity than the
cleanup is worth.

Tests: 186/186 pass. Skills + site build clean. _headers,
_redirects, _routes.json, _data/ all land in build/ correctly.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(cli): consolidate bin/ + src/ + lib/ under cli/

Phase 2 step 5 of the directory restructure. The CLI surface was split
across three top-level dirs whose names were easy to mistake for each
other (especially src/ vs source/ pre-step-2). Consolidates under cli/.

- git mv bin -> cli/bin (CLI entry + skills sub-command)
- git mv src -> cli/engine (detect-antipatterns engine + browser variant)
- git mv lib -> cli/lib (download-providers helper)

Update package.json:
- bin.impeccable: cli/bin/cli.js
- main + exports: cli/engine/detect-antipatterns.mjs and the
  ./browser variant
- files: ["cli/", "LICENSE"]

Update internal references:
- cli/bin/cli.js: dynamic import points at ../engine/, package.json
  read goes one level deeper (../../package.json).
- functions/api/download/[type]/[provider]/[id].js + bundle/[provider].js:
  cli/lib/download-providers.js path.
- scripts/build.js, scripts/build-browser-detector.js,
  scripts/build-extension.js: cli/engine path constants.
- scripts/lib/sub-pages-data.js, scripts/lib/utils.js, skill/scripts/
  live-server.mjs: comment refs.
- tests/detect-antipatterns{,-browser,-fixtures}.test.{js,mjs},
  tests/windows-path-fix.test.js: import + read paths.
- AGENTS.md, CLAUDE.md: doc paths.

Verified:
- npx node cli/bin/cli.js --version, --help, detect --help all work.
- bun run build, bun run build:browser, bun run build:extension all
  clean. Browser detector lands at cli/engine/detect-antipatterns-browser.js;
  extension/detector/detect.js still emits to the same location.
- bun run test: 186/186 pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: update browser-detector paths missed in cli/ rename

Bugbot caught two runtime path leaks where the comment got renamed
to cli/engine/ but the actual code still used the old src/ segment.

- skill/scripts/live-server.mjs: detectPaths array now joins cli, engine,
  detect-antipatterns-browser.js for both the repo-relative lookup
  (4 dirs up from .claude/skills/impeccable/scripts/ to repo root) and
  the npm node_modules fallback. Without this fix, the detection
  overlay would silently not load during live-server sessions.

- scripts/build.js: the post-build copy of the browser detector into
  site/public/js/ was reading from src/. The if (fs.existsSync(...))
  guard meant the copy was silently skipping, so antipattern-examples
  pages would 404 on /js/detect-antipatterns-browser.js once the site
  was deployed.

Tests: 186/186 pass. Build clean. site/public/js/detect-antipatterns-browser.js
re-emits as expected.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: cleanup-deprecated import path missed an extra .. in cli/ rename

Bugbot caught three call sites in cli/bin/commands/skills.mjs that
import '../../skill/scripts/cleanup-deprecated.mjs'. Pre-rename, that
was correct from bin/commands/ (one parent to bin/, one to repo root).
After moving the file from bin/commands/ to cli/bin/commands/, the
path is one directory deeper, so it needs three .. segments to reach
the repo root. Without the fix, every cleanup invocation throws on
import and gets swallowed by the surrounding try/catch — silent skip.

cli/bin/cli.js's package.json read already uses '../../package.json'
(the same depth pattern), confirming three levels is correct.

Verified: dynamic import resolves and exports the expected functions.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* chore: sweep stale path/file references missed in the restructure

Same root cause as the two bugbot finds: some references in moved or
related files weren't tracked because they didn't match a simple
sed pattern. Caught the rest by walking each moved dir's depth and
each Astro-migration deletion.

Stale path references (post-Astro migration, missed earlier):
- CLAUDE.md: legacy URL redirects "live in server/index.js" -> point
  at the actual sources (scripts/build.js generateCFConfig +
  site/public/_redirects).
- AGENTS.md: counts.js path (public/ -> site/public/), changelog file
  (public/index.html -> site/pages/index.astro), screenshots note
  (public/ -> site/), source-of-truth dirs (source/, src/ -> skill/,
  cli/).
- tests/detect-antipatterns-browser.test.mjs: comment about routes
  "in server/index.js".
- skill/reference/live.md: workflow.css example for "this repo" was
  pre-Astro (public/css/) -> site/styles/. (User-project Vite/Next
  example unchanged.)

Stale path that pointed at moved files:
- tests/skills-cli.test.js: CLI path was '..', 'bin', 'cli.js'; now
  '..', 'cli', 'bin', 'cli.js'. Test isn't wired into bun run test
  but it would have failed if invoked.

Dead files (orphaned by Astro migration, never cleaned up):
- tests/server/download-validation.test.js: imported from
  ../../server/lib/{validation,api-handlers}.js which were deleted in
  b8f09c8. Test was a silent failure waiting to happen.
- scripts/lib/render-markdown.js: 156-line module with zero consumers
  (the only caller, scripts/lib/render-page.js, was deleted in the
  Astro cleanup).
- scripts/build.js: dead commented-out generateSubPages import.

Tests: 186/186 pass. Build clean.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(build): remove invalid Corepack packageManager spec

Cloudflare Pages rejects the build with `Unsupported package manager
specification (bun@1.3.11)`. The packageManager field follows
Corepack's syntax which only validates npm/pnpm/yarn — `bun@X.Y.Z`
parses as a malformed Corepack directive even though Bun itself
treats it as a hint.

Pre-existing on main since d874af0 (CF Pages deploy on main also
failing); just surfaces here because the PR triggers a fresh deploy.

CF Pages auto-detects Bun anyway (the build log confirms:
"Detected the following tools from environment: bun@1.3.11,
pnpm@10.11.1, nodejs@22.16.0"). Removing the field unblocks the
deploy without changing local dev behavior.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Paul Bakaus <paulbakaus@pauls-mbp-3.lan>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Paul Bakaus
2026-05-04 16:38:03 -07:00
committed by GitHub
co-authored by Paul Bakaus Claude Opus 4.7
parent 2aeac48b19
commit e587004ee4
256 changed files with 383 additions and 2650 deletions
+269
View File
@@ -0,0 +1,269 @@
# ADR: Live Variant Mode
**Status:** Implemented (v3.1, feature branch `feature/single-skill-consolidation`)
**Date:** 2026-04-12
**Author:** Paul Bakaus + Claude
## Context
Impeccable is a design skill for AI coding agents. It teaches AI harnesses (Claude Code, Cursor, Gemini CLI, Codex, etc.) how to produce better frontend design. The skill has 22 commands (bolder, quieter, polish, typeset, etc.) that the agent runs on source code.
The missing piece: there was no way to visually iterate on a live page. The user could ask the agent to "make this bolder," but they had to read the code diff, reload the page, and decide if they liked it. If not, they'd ask again, wait, reload, repeat. Slow and disconnected.
**Goal:** Let the user select an element directly in the browser, pick a design action, and see N real HTML+CSS variants hot-swapped in. Cycle through them visually, accept or discard, repeat. The agent generates the variants; the browser shows them.
## Decision
Build a self-contained live variant mode that ships as part of the impeccable skill (no separate npm install required). The system bridges three parties: the **browser** (where the user picks elements and cycles variants), the **server** (a localhost HTTP server that relays messages), and the **agent** (the AI that generates variants by modifying source files).
### Key architectural decisions
**1. Source modification, not DOM patching.**
Variants are written to the actual source file, not injected into the browser DOM. This means:
- Framework state (React, Vue, etc.) is preserved because the framework's own rendering pipeline handles the update via HMR.
- "Accept" is trivial: the winning variant is already in the source. Just remove the other variants.
- Variants are real code that the user can inspect in their editor, diff, and commit.
**2. SSE + fetch, not WebSocket.**
Server-Sent Events (server to browser) + fetch POST (browser to server) instead of WebSocket. This eliminates the `ws` npm dependency entirely. The server is zero-dependency pure Node.js (http, crypto, fs, net, os). This matters because the scripts ship inside the skill directory and run in the user's project without any package installation.
**3. Self-contained skill scripts.**
All live mode code lives in `skill/scripts/`:
- `live-server.mjs` — HTTP server (SSE, poll, source file reader)
- `live-poll.mjs` — CLI client for the agent poll/reply loop
- `live-wrap.mjs` — CLI helper that finds elements in source and creates variant wrappers
- `live-browser.js` — Browser script (element picker, action panel, variant cycler, global bar)
When a user installs the skill via `npx skills add pbakaus/impeccable`, they get the live mode without any additional setup. The agent runs the scripts via `node {{scripts_path}}/live-server.mjs`.
**4. HTTP long-poll for the agent, not WebSocket or stdin.**
The agent communicates with the server via HTTP long-poll (`GET /poll` blocks until a browser event arrives). This works across all AI harnesses because every harness can run a shell command and read its stdout. No harness-specific integration needed.
**5. `display: contents` variant wrapper.**
Variants are wrapped in a container with `display: contents`, which makes the wrapper invisible to CSS layout. The selected element's relationship with its parent (flex child, grid child, etc.) is preserved. The wrapper carries `data-impeccable-variants` and `data-impeccable-variant-count` attributes that the browser script uses to detect and cycle variants.
**6. No-HMR fallback.**
For dev servers that don't support HMR (like Bun's static HTML import), the browser fetches the raw source file directly from the live server's `/source` endpoint and injects the variants into the DOM. This works universally, at the cost of losing framework state on that injection.
## Architecture
```
┌─────────────────────────────────────────────────────────────────┐
│ BROWSER │
│ │
│ live-browser.js (injected via <script> tag in source HTML) │
│ ├── Element picker (mousemove highlight, click select, │
│ │ keyboard nav: arrows=siblings, shift+arrows=parent/child) │
│ ├── Action bar (floating pill: action picker, freeform input, │
│ │ variant count, go button; morphs to generating/cycling) │
│ ├── Global bar (bottom pill: Detect toggle, Pick toggle, Exit) │
│ ├── Variant cycler (MutationObserver watches for new │
│ │ [data-impeccable-variant] children in DOM) │
│ ├── SSE connection (EventSource → /events for server push) │
│ ├── fetch POST (→ /events for browser-to-server messages) │
│ └── localStorage (session state survives reloads) │
│ │
└────────────┬──────────────────────────────────┬─────────────────┘
│ SSE (server → browser) │ POST (browser → server)
│ EventSource /events │ fetch /events
▼ ▼
┌─────────────────────────────────────────────────────────────────┐
│ LIVE SERVER │
│ │
│ live-server.mjs (node, localhost:8400+, zero dependencies) │
│ ├── GET /live.js — browser script (token injected) │
│ ├── GET /detect.js — anti-pattern overlay (backwards compat)│
│ ├── GET /events — SSE stream to browser (server push) │
│ ├── POST /events — browser events (generate, accept, etc.)│
│ ├── GET /poll — agent long-poll (blocks until event) │
│ ├── POST /poll — agent reply (forwarded to browser SSE) │
│ ├── GET /source — raw file reader (no-HMR fallback) │
│ ├── GET /health — status, port, connected clients │
│ └── GET /stop — graceful shutdown │
│ │
│ State: session token, SSE client set, event queue, poll queue │
│ Server file: .impeccable/live/server.json (project root) │
│ │
└────────────┬──────────────────────────────────┬─────────────────┘
│ GET /poll (long-poll) │ POST /poll (reply)
│ blocks until browser event │ forwarded to SSE
▼ ▲
┌─────────────────────────────────────────────────────────────────┐
│ AGENT │
│ │
│ Follows skill/reference/live.md │
│ 1. Start server: node scripts_path/live-server.mjs & │
│ 2. Inject <script> into source HTML (comment-marked) │
│ 3. Poll loop: │
│ ├── generate → wrap + write variants + reply done │
│ ├── accept → present variant code + cleanup + reply done │
│ ├── discard → restore original + reply done │
│ ├── exit → cleanup script tag + stop server │
│ └── timeout → re-poll │
│ │
│ Tools used per generation: │
│ 1. node live-wrap.mjs (find element, create wrapper) │
│ 2. Edit (write all variants in single edit) │
│ 3. node live-poll.mjs --reply <id> done --file <path> │
│ │
└─────────────────────────────────────────────────────────────────┘
```
## Message flow
### Generate variants
```
User clicks element → picks "Bolder" → clicks Go
Browser POST /events: {type:"generate", id:"abc", action:"bolder", count:3, element:{...}}
Server enqueues event
Agent GET /poll returns: {type:"generate", id:"abc", action:"bolder", count:3, element:{...}}
Agent runs: node live-wrap.mjs --id abc --count 3 --classes "hero-left"
→ Finds element in source, wraps with data-impeccable-variants container
→ Original stays visible (no flash of empty content)
Agent writes all 3 variants in a single Edit tool call
→ Each variant is a <div data-impeccable-variant="N"> with full HTML replacement
→ First variant visible, others display:none
Agent POST /poll: {id:"abc", type:"done", file:"public/index.html"}
Server forwards via SSE to browser: {type:"done", file:"public/index.html"}
Browser checks: variants in DOM? (HMR)
YES → MutationObserver detected them → show cycling bar
NO → fetch /source?path=public/index.html → parse → inject → show cycling bar
```
### Accept variant
```
User clicks Accept on variant 2
Browser POST /events: {type:"accept", id:"abc", variantId:"2"}
Browser shows "Applying variant..." spinner
Agent GET /poll returns: {type:"accept", id:"abc", variantId:"2"}
Agent reads variant 2 HTML from source, presents to user
Agent removes variant wrapper, restores clean source
Agent POST /poll: {id:"abc", type:"done"}
Browser receives done via SSE → green "Variant applied" confirmation → auto-dismiss
```
### Discard
```
User clicks Discard (or presses Escape)
Browser POST /events: {type:"discard", id:"abc"}
Browser immediately: restores original element in DOM, hides bar, resets to PICKING
Agent GET /poll returns: {type:"discard", id:"abc"}
Agent removes variant wrapper from source, restores original
Agent POST /poll: {id:"abc", type:"done"}
```
## Variant wrapper format
In the source file (HTML example):
```html
<!-- impeccable-variants-start abc12345 -->
<div data-impeccable-variants="abc12345" data-impeccable-variant-count="3" style="display: contents">
<div data-impeccable-variant="original">
<!-- original element (visible until first variant arrives) -->
</div>
<div data-impeccable-variant="1">
<!-- variant 1 (visible) -->
</div>
<div data-impeccable-variant="2" style="display: none">
<!-- variant 2 -->
</div>
<div data-impeccable-variant="3" style="display: none">
<!-- variant 3 -->
</div>
</div>
<!-- impeccable-variants-end abc12345 -->
```
Comment markers enable deterministic cleanup. `display: contents` on the wrapper preserves flex/grid layout. The `data-impeccable-variant-count` attribute tells the browser how many to expect.
## Browser UI
### Global bar (always visible during live mode)
Compact floating pill at bottom center. Light, translucent, matching the brand aesthetic.
- **Impeccable** brand mark (magenta)
- **Detect** toggle (eye icon): loads anti-pattern scanner in extension mode, shows overlay count badge
- **Pick** toggle (crosshair icon): enables/disables element picker (default: on)
- **Exit** button (x): sends exit event, tears down all UI
When both Detect and Pick are active, detect overlays get `pointer-events: none` so the picker sees through them. The picker's z-index (100001) is above detect overlays (99999).
### Action bar (floating, contextual)
Appears below the selected element. Morphs between states:
- **Configure**: `[Action pill ▾] [freeform input] [×3] [Go →]`
- **Generating**: `[Action label] [● ● ○] Generating 2 of 3...`
- **Cycling**: `[←] [● ● ●] 2/3 [→] [✓ Accept] [✕]`
- **Saving**: `[spinner] Applying variant...`
- **Confirmed**: `[✓ Variant applied]` (green, auto-dismisses after 1.8s)
## Session persistence
`localStorage` stores:
- Session state (id, action, count, arrived variants, visible variant)
- Handled sessions (accepted/discarded session IDs)
This survives page reloads, browser close/reopen, HMR, and accidental refreshes. On page load, `resumeSession()` checks for an active variant wrapper in the DOM + the localStorage state and resumes the correct cycling position.
## Security
- **Session token**: `crypto.randomUUID()`, checked on all mutating endpoints and SSE connections.
- **Localhost only**: server binds to `127.0.0.1`, not `0.0.0.0`.
- **Token in server file**: `.impeccable/live/server.json` in project root. Only the user's processes can read it.
- **Token injected into `/live.js`**: the server prepends `window.__IMPECCABLE_TOKEN__` at serve time.
- **Path traversal guard**: `/source` endpoint validates the requested path is within `process.cwd()`.
- **No eval/innerHTML**: all browser UI built with `createElement` and `textContent`.
## Server resilience
- **Debounced exit**: when all SSE clients disconnect, the server waits 8 seconds before signaling exit to the agent. This avoids false exits from HMR reloads and brief network blips.
- **Stale PID detection**: on startup, the server checks if an existing PID file's process is still running. Dead processes are cleaned up automatically.
- **Browser server-lost handling**: after 5 failed SSE reconnection attempts, the browser cleans up all UI and shows a "Live server disconnected" toast.
## Performance optimizations
The generation loop was optimized from ~40s to ~15-20s:
1. **`wrap` CLI helper**: one command replaces 3-4 agent tool calls (grep + read + edit). Finds the element by ID/class/tag priority, creates the variant wrapper, returns the file path and insert line.
2. **Batch variant writes**: all variants in a single file edit instead of one per variant. Saves N-1 tool call round-trips.
3. **Page URL hint**: browser includes `location.pathname` in the generate event so the agent can map URL to source file directly.
Net effect: 4 tool calls (wrap + edit + read + reply) instead of 8+.
## Test coverage
- **26 wrap tests** (`tests/live-wrap.test.mjs`): unit tests for `buildSearchQueries`, `findElement`, `findClosingLine`, `detectCommentSyntax` + integration tests for full `wrapCli` on HTML and JSX fixtures.
- **15 server tests** (`tests/live-server.test.mjs`): integration tests that start a real server, test all endpoints, verify browser→agent event flow and agent→browser SSE delivery.
## Known limitations
- **Bun's static HTML import**: Bun's `import from "index.html"` caches at module load time. Source changes require a server restart to appear in the served HTML. The no-HMR fallback (fetch from `/source`) handles this, but it's less seamless than Vite/Next.js where HMR works natively.
- **Single generation at a time**: only one generate/cycle session can be active. This is by design (the source file can only have one variant wrapper at a time).
- **Inspection-only accept (v1)**: accepting a variant presents the code to the user and cleans up the scaffolding. Write-back (keeping the variant in place) is planned for v2.
- **No cancel during generation**: once the agent starts generating, it finishes all variants before the user can interact again.
## Future work
- **Write-back on accept**: instead of restoring the original after accept, keep the accepted variant as the new source.
- **Detect-to-fix flow**: clicking a detected anti-pattern pre-selects that element and pre-fills the action (e.g., "overused font" pre-fills "typeset").
- **Background poll for Claude Code**: use `run_in_background: true` to keep the main conversation free during the poll loop.
- **Streaming variants**: progressive reveal as the agent writes each variant (currently batched for speed).
@@ -0,0 +1,555 @@
---
title: feat: Add durable live session recovery
type: feat
status: active
date: 2026-04-28
deepened: 2026-04-28
---
# feat: Add durable live session recovery
## Overview
Impeccable live mode should remain recoverable when the browser changes live-session state while the agent is not actively polling, after the agent is interrupted, or after the helper server/page reloads. The change adds a durable live-session journal, a resumable server/browser state contract, and an agent-facing status command so a fresh agent can reconstruct the active session and continue from the correct next action.
This plan does not change the core source-first live-mode architecture from `docs/adr-live-variant-mode.md`: variants are still written to source, the browser still previews through HMR or `/source`, and the agent still uses long-poll. It makes the state machine durable and inspectable instead of relying on chat memory plus in-process queues.
---
## Problem Frame
The observed failure mode was: the user accepted and tuned a live variant in the browser while the agent was not listening continuously. The browser moved forward, source entered a transitional state, and the agent no longer had authoritative context for what the user had done. Current code already has useful browser `localStorage` resume behavior and server in-memory queues, but neither is sufficient as a recovery source after interruption, server restart, browser cleanup, or missed polling.
The product expectation is stronger: if the user changes live-mode state in the browser, a later agent should be able to ask, “what happened and what do I need to do next?” and receive a complete, durable answer.
---
## Requirements Trace
- R1. Browser live-session changes are durably recorded outside browser memory and chat context.
- R2. A new or interrupted agent can reconstruct the current live session, selected variant, parameter values, source file, and next required action.
- R3. Accept/discard events are not silently lost when the agent is not polling or when the helper server restarts.
- R4. Browser resume behavior treats source markers and durable session state as recoverable truth rather than hiding unfinished work behind a local handled flag.
- R5. Durable replay is idempotent: duplicate, late, or conflicting events must not corrupt source.
- R6. Carbonize-required accepts are represented as an explicit incomplete state until cleanup is done, and can reconcile to complete when source markers prove cleanup already happened.
- R7. Existing browser/agent event transport remains self-contained, zero-dependency, token-protected, and compatible with the current SSE plus long-poll protocol. Status/resume may be implemented as a separate CLI/file-reader API.
- R8. Agent delivery uses an explicit lease/ack model so a poll response is not treated as completed work until the agent reports a result.
- R9. Annotated screenshots and generated-file fallback metadata remain recoverable while a session is incomplete.
---
## Scope Boundaries
- This plan does not replace long-poll with WebSockets or a harness-specific integration.
- This plan does not make accept fully atomic in one step. It records and resumes the incomplete state first; a later plan may remove carbonize as a manual cleanup stage.
- This plan does not redesign generated variant quality or CSS specificity behavior, except where preview state metadata needs to be captured.
- This plan does not introduce a database or external service. Durable state should live in project-local files managed by the helper server.
- This plan does not support multiple simultaneous browser tabs editing different sessions as a first-class collaboration mode. It should avoid corrupting them, but single active session remains the baseline.
### Deferred to Follow-Up Work
- Atomic accept/carbonize: a later plan can make `live-accept.mjs` produce permanent clean source in one deterministic step.
- Safer generated-preview layout defaults: a later plan can harden wrapper sizing, overflow, and scoped selector guidance.
- Multi-user or multi-tab collaboration semantics: out of scope until live mode intentionally supports collaborative sessions.
---
## Context & Research
### Relevant Code and Patterns
- `skill/scripts/live-server.mjs` owns `/events`, `/poll`, `/source`, `/health`, token validation, in-memory `pendingEvents`, and browser SSE clients.
- `skill/scripts/live-browser.js` owns picker state, variant cycling, parameter controls, `localStorage` session resume, handled-session sentinels, and accept/discard browser behavior.
- `skill/scripts/live-poll.mjs` is the agent-facing poll client and auto-runs `live-accept.mjs` for accept/discard events.
- `skill/scripts/live-accept.mjs` deterministically accepts/discards variant wrappers and can emit carbonize-required results.
- `skill/scripts/live-wrap.mjs` creates source markers and original/variant wrapper structure.
- `tests/live-server.test.mjs`, `tests/live-accept.test.mjs`, `tests/live-wrap.test.mjs`, and `tests/live-e2e.test.mjs` are the relevant verification surfaces.
- `docs/adr-live-variant-mode.md` documents the current architecture and should be updated if the durable journal changes the lifecycle contract.
### Institutional Learnings
- `docs/adr-live-variant-mode.md` explicitly values source modification over DOM patching, zero-dependency scripts, SSE plus fetch, long-poll for agent compatibility, and `display: contents` wrappers.
- `skill/reference/live.md` currently encodes the operational assumption that the agent continuously polls and performs carbonize cleanup before the next poll.
### External References
- External research skipped. This is a repo-owned local protocol and the existing architecture is well documented; local patterns are more authoritative than generic event-sourcing guidance.
---
## Key Technical Decisions
- Add a project-local durable live-session store rather than relying on browser `localStorage` or server memory alone: this is the minimum change that lets any future agent reconstruct state.
- Use append-only session events plus a compact session snapshot: append-only events preserve the audit trail for recovery; the snapshot makes status reads fast and simple. The append-only journal and source markers are canonical; snapshots are rebuildable caches.
- Treat poll delivery as a leased work item, not queue removal: an event becomes terminal only after the agent posts a result/ack, and stale leases are redelivered idempotently.
- Keep browser `localStorage` as a UI convenience, not the source of truth: source markers plus durable server journal should win when they disagree.
- Add explicit server-side session states instead of ad hoc flags: states make accept/discard replay, stale checkpoint rejection, and invalid transitions enforceable.
- Make accept/discard delivery acknowledged before the browser clears recoverable state: this prevents “browser thinks handled, source did not change” split-brain.
- Add agent-facing status/resume scripts instead of requiring agents to inspect raw files: live mode needs a stable agent API, not folklore.
---
## Open Questions
### Resolved During Planning
- Should the first fix be durable resumability or atomic accept? Durable resumability comes first because it solves missed polling, interrupted agents, and transitional source states without requiring a full accept rewrite.
- Should durability depend on external storage? No. Live mode is intentionally self-contained and should use project-local files.
### Deferred to Implementation
- Exact file naming inside the session store: implementation should choose a simple structure after reviewing how `.impeccable-live.json` is currently managed.
- Exact retention policy defaults: implementation should start conservative and prune old completed sessions only after tests cover recovery.
- Exact browser UI copy for pending accept/reconnect states: copy should be short and can be refined during implementation.
- Exact helper-server restart UX: auto-rebind across token/port rotation is out of scope for this plan unless implementation discovers a safe lightweight path. The default recovery is `live-status` plus asking the user to reload/reopen the instrumented page when needed.
---
## Output Structure
skill/scripts/
live-session-store.mjs
live-status.mjs
live-resume.mjs
live-server.mjs
live-browser.js
live-poll.mjs
live-accept.mjs
tests/
live-session-store.test.mjs
live-server.test.mjs
live-browser-recovery.test.mjs
live-poll.test.mjs
live-status.test.mjs
live-e2e.test.mjs
package.json
docs/
adr-live-variant-mode.md
---
## High-Level Technical Design
> *This illustrates the intended approach and is directional guidance for review, not implementation specification. The implementing agent should treat it as context, not code to reproduce.*
```mermaid
stateDiagram-v2
[*] --> configuring
configuring --> generating: browser Go
generating --> source_wrapped: agent wrapped source; file known
source_wrapped --> variants_written: agent wrote variants
variants_written --> cycling: agent replies done or browser sees wrapper
cycling --> accept_pending: browser Accept
cycling --> discard_pending: browser Discard
accept_pending --> accepted_source_pending: agent/live-accept wrote accepted variant with carbonize markers
accepted_source_pending --> complete: carbonize cleanup done
discard_pending --> complete: source restored
generating --> stranded: server/page/agent interruption
cycling --> stranded: server/page/agent interruption
accept_pending --> stranded: server/page/agent interruption
stranded --> generating: resume pending leased event
stranded --> source_wrapped: resume wrapper/original-only source
stranded --> cycling: resume from durable journal + source markers
stranded --> accepted_source_pending: resume from carbonize markers
```
```mermaid
sequenceDiagram
participant B as Browser
participant S as Live server
participant J as Session journal
participant A as Agent
participant F as Source files
B->>S: POST /events accept_intent(session, variant, params)
S->>J: append event; rebuild/atomically update snapshot(accept_pending)
S-->>B: 202 accepted + durable sequence
A->>S: GET /poll or live-status
S->>J: lease next pending event
S-->>A: accept event + recovery context + lease id
A->>F: live-accept / cleanup
A->>S: POST /poll done + file/result + lease id
S->>J: append agent_result; mark lease acked; update snapshot
S-->>B: SSE done/committed
B->>B: clear local pending state only after committed
```
---
## Implementation Units
- U1. **Define durable session store**
**Goal:** Add a small storage module that can append events, update snapshots, read active sessions, and recover pending work after process restart.
**Requirements:** R1, R2, R3, R5, R7, R8, R9
**Dependencies:** None
**Files:**
- Create: `skill/scripts/live-session-store.mjs`
- Create: `tests/live-session-store.test.mjs`
- Modify: `skill/scripts/live-server.mjs`
- Modify: `package.json`
**Approach:**
- Store state under a project-local live directory, separate from `.impeccable-live.json` but discoverable from the same project root.
- Write append-only JSONL events per session and a compact JSON snapshot per session. The JSONL journal is authoritative; snapshots are derived and must be rebuildable when missing, stale, or contradictory.
- Include session id, event sequence, event type, phase, source file when known, visible variant, param values, timestamps, delivery lease/ack metadata, checkpoint revision, generated-file fallback mode, and annotation artifact paths.
- Persist annotation screenshot assets for incomplete sessions when generate events include annotations; status should report missing artifacts as diagnostics.
- Keep the module dependency-free and synchronous or simple async Node filesystem code consistent with current scripts.
- Route browser checkpoints through `/events` as a first-class `checkpoint` event type, with monotonic client revision and session owner metadata.
- Wire new non-E2E tests into the default test script, because `package.json` enumerates test files explicitly rather than globbing all `tests/*.test.mjs`.
- Treat malformed journal entries as recoverable diagnostics rather than crashing the helper server when possible.
**Execution note:** Start test-first for the storage module because it is the new source of truth.
**Technical design:** Directional event shape:
```text
SessionSnapshot = {
id,
phase,
pageUrl,
sourceFile,
expectedVariants,
arrivedVariants,
visibleVariant,
paramValues,
pendingEventSeq,
deliveryLease,
checkpointRevision,
activeOwner,
sourceMarkers,
fallbackMode,
annotationArtifacts,
diagnostics,
updatedAt
}
```
**Patterns to follow:**
- `skill/scripts/live-server.mjs` for project-root PID file handling.
- `tests/live-server.test.mjs` for temp-directory test isolation.
**Test scenarios:**
- Happy path: appending `generate`, `variants_ready`, and `accept_intent` events for one session yields a snapshot with the latest phase and selected variant.
- Happy path: reading active sessions after constructing a new store instance returns the session written by the previous store instance.
- Edge case: duplicate event id or sequence is idempotent and does not append conflicting state twice.
- Edge case: a completed session remains readable for audit but is not returned as the active session by default.
- Edge case: journal and snapshot disagree after a simulated crash; status rebuilds the snapshot from the journal and records a repair diagnostic.
- Edge case: annotated generate event keeps its screenshot artifact available until the session completes.
- Error path: corrupted JSONL line is reported in diagnostics while valid prior events still reconstruct the snapshot.
- Error path: storage directory creation failure returns a structured error to the caller.
**Verification:**
- Durable state survives module re-instantiation and can answer “what is the active session and next action?” without browser or server memory.
---
- U2. **Journal browser events and enforce session transitions**
**Goal:** Make `/events` persist generate/accept/discard and checkpoint events before they are exposed to agent polling.
**Requirements:** R1, R3, R5, R7, R8
**Dependencies:** U1
**Files:**
- Modify: `skill/scripts/live-server.mjs`
- Modify: `tests/live-server.test.mjs`
**Approach:**
- On `POST /events`, validate generate/accept/discard/checkpoint events, assign or validate a durable event sequence, append to the session journal, then enqueue actionable events for polling. Checkpoints update durable state but do not necessarily create agent work.
- Add a server-side session state machine that accepts valid transitions and treats duplicate valid events idempotently.
- Add an at-least-once poll delivery model: pending events are leased to a poll response, redelivered after lease expiry, and marked complete only when the agent posts a matching result.
- Keep `exit` as lower priority than real session events so queued generate/accept/discard work is not masked by tab disconnect.
- Preserve current token validation and in-memory fast path, but make disk the replayable source.
- On server startup, rebuild pending work from the journal into the in-memory queue. `/poll` may consult the store when memory is empty, but the journal remains canonical.
**Patterns to follow:**
- Existing `validateEvent()` and `enqueueEvent()` in `skill/scripts/live-server.mjs`.
- Existing `/events` and `/poll` tests in `tests/live-server.test.mjs`.
**Test scenarios:**
- Happy path: `POST /events` for generate persists an event before a poll consumes it.
- Happy path: if the server object is recreated with the same project store, `/poll` returns the previously unconsumed generate event.
- Edge case: duplicate accept for the same session and variant returns the same durable state without creating conflicting queue entries.
- Edge case: discard after accept-pending is rejected or ignored according to the state machine and returns a diagnostic response.
- Error path: invalid transition returns a clear JSON error and does not append to the journal.
- Integration: queued real events are delivered before synthetic `exit` events.
- Integration: event delivered to a poll but never acked is redelivered after lease timeout and remains idempotent when eventually completed.
**Verification:**
- Browser events are not lost by an idle agent or helper server restart as long as the project-local journal remains.
---
- U3. **Add browser checkpoints and acknowledged accept/discard**
**Goal:** Have the browser record current live state durably and stop clearing recoverable local state until the server acknowledges the state change.
**Requirements:** R1, R2, R3, R4, R5
**Dependencies:** U1, U2
**Files:**
- Modify: `skill/scripts/live-browser.js`
- Modify: `tests/live-e2e.test.mjs`
- Create: `tests/live-browser-recovery.test.mjs`
**Approach:**
- Add a browser checkpoint helper that sends current session state to `/events` as `type: "checkpoint"` whenever variant, parameter values, phase, or selected source file context changes.
- Include monotonic client revisions and an active-session owner/epoch so stale checkpoints from old tabs cannot regress a newer accepted/discarded phase.
- Change accept/discard from fire-and-forget to acknowledged durable receipt before `markSessionHandled()` and before local session cleanup. Agent completion remains a later state.
- Keep a local pending state if acknowledgement fails, with a visible “waiting for agent/server” or “reconnect to recover” state instead of silently resetting.
- Preserve existing `localStorage` resume as a UI fast path, but do not let its single handled key suppress recovery when source markers or durable server state indicate pending work.
- Capture current parameter values on every change, not only at accept time, so resume can reconstruct user tuning even before Accept.
**Patterns to follow:**
- Existing `saveSession()`, `resumeSession()`, `paramsCurrentValues`, and `handleAccept()` in `skill/scripts/live-browser.js`.
- Existing scroll restoration and MutationObserver recovery patterns in `skill/scripts/live-browser.js`.
**Test scenarios:**
- Happy path: moving a parameter slider sends a checkpoint with updated param values and does not reset the tune panel.
- Happy path: clicking Accept receives server acknowledgement, then transitions the browser to confirmed/pending-agent state.
- Edge case: Accept POST fails due to server down; browser keeps session recoverable and shows a reconnect/retry state.
- Edge case: page reload after accept acknowledgement but before agent completion resumes as pending cleanup rather than picking mode.
- Edge case: local handled flag exists but source wrapper still exists; browser resumes or reports pending cleanup instead of hiding the session.
- Edge case: stale checkpoint from a background tab arrives after accept-pending; server rejects it or records it as stale without changing snapshot phase.
- Integration: after HMR inserts variants, browser checkpoint records arrived variant count and visible variant.
**Verification:**
- The browser never becomes the only place where selected variant and tune values exist.
---
- U4. **Expose agent status and resume commands**
**Goal:** Give agents a stable CLI/API to inspect and continue live state without manually reading raw journal files or source markers.
**Requirements:** R2, R4, R6, R7
**Dependencies:** U1, U2
**Files:**
- Create: `skill/scripts/live-status.mjs`
- Create: `skill/scripts/live-resume.mjs`
- Create: `skill/scripts/live-complete.mjs`
- Modify: `skill/scripts/live-server.mjs`
- Modify: `skill/reference/live.md`
- Modify: `tests/live-server.test.mjs`
- Create: `tests/live-status.test.mjs`
**Approach:**
- Add a status endpoint or direct store reader that returns active session snapshots, pending events, source marker status, and recommended next action.
- Add `live-status.mjs` for human/agent-readable JSON status.
- Add `live-resume.mjs` if the status command should also requeue pending events or print a normalized next event for the agent.
- Include source-marker scanning for `impeccable-variants-start`, `impeccable-carbonize-start`, and `impeccable-param-values` so source truth can repair journal/browser drift. The scan domain should include resolved live config targets, journal-known files, and existing source roots used by marker helpers; generated/served files covered by live config must be diagnosable.
- Consult the generated-file guard before recommending deterministic `live-accept`; generated/served-file fallback sessions should return `persist_fallback_to_true_source`.
- Distinguish generate recovery phases: pending event only, wrapper/original-only source, variants partially/fully written, and browser not yet confirmed cycling.
- Keep output JSON stable and agent-readable: explicit `nextAction`, `reason`, `file`, `sessionId`, `paramValues`, `lease`, and `diagnostics` fields.
- After completion acknowledgement, completed sessions return `no_active_session` by default unless an include-completed flag is requested.
**Technical design:** Directional `nextAction` values:
```text
poll_for_pending_event
write_variants
continue_writing_variants
run_accept_cleanup
run_carbonize_cleanup
acknowledge_completion
persist_fallback_to_true_source
restore_discard
ask_user_to_reopen_browser
no_active_session
```
**Patterns to follow:**
- `skill/scripts/live-poll.mjs` for CLI JSON output style.
- `skill/scripts/live-accept.mjs` marker parsing helpers where reusable.
**Test scenarios:**
- Happy path: status with a pending accept event returns `run_accept_cleanup` and includes variant id and param values.
- Happy path: status with carbonize markers in source returns `run_carbonize_cleanup` even if no browser is connected.
- Edge case: status with no active journal but source variant markers returns a recoverable source-marker session.
- Edge case: status with stale completed sessions returns `no_active_session` by default and can include completed sessions only when requested.
- Edge case: source markers exist in a generated/served file covered by live config; status recommends fallback true-source persistence rather than deterministic accept.
- Edge case: wrapper exists with only original content; status recommends continuing variant writing rather than accept cleanup.
- Error path: missing or unreadable source file appears as a structured diagnostic, not an unhandled exception.
- Integration: `live-resume.mjs` can requeue or emit the pending event after a helper server restart.
**Verification:**
- A fresh agent can run one command and know exactly what happened in live mode and what to do next.
---
- U5. **Make poll/accept acknowledge durable completion**
**Goal:** Ensure agent-side handling updates durable session state after source writes and carbonize cleanup so browser and future agents know whether the session is complete.
**Requirements:** R2, R5, R6
**Dependencies:** U1, U2, U4
**Files:**
- Modify: `skill/scripts/live-poll.mjs`
- Modify: `skill/scripts/live-accept.mjs`
- Modify: `skill/scripts/live-server.mjs`
- Modify: `tests/live-accept.test.mjs`
- Modify: `tests/live-server.test.mjs`
- Create: `tests/live-poll.test.mjs`
**Approach:**
- Have `live-poll.mjs` report auto-accept/discard results back to the server/session store, not only print `_acceptResult` to stdout.
- Complete the delivery lease only after this result is durably recorded; failed source writes should release or mark the lease recoverable with diagnostics.
- Represent `carbonize: true` as `accepted_source_pending` until cleanup is confirmed.
- Add `live-complete.mjs` as the canonical completion acknowledgement path after carbonize cleanup. `live-status.mjs` reports that completion is needed; `live-resume.mjs` helps recover pending work; only `live-complete.mjs` marks carbonize cleanup complete.
- Reconcile lost completion acknowledgements from source truth: if carbonize markers are gone and accepted content is materialized, status can idempotently record a synthetic completion event.
- Make duplicate completion acknowledgements idempotent.
- Preserve the stderr warning as a human attention signal, but do not rely on warning text as the state machine.
**Patterns to follow:**
- Current `_acceptResult` attachment in `skill/scripts/live-poll.mjs`.
- Current carbonize marker output in `skill/scripts/live-accept.mjs`.
**Test scenarios:**
- Happy path: accept event processed by `live-poll.mjs` updates durable session to `accepted_source_pending` when carbonize is required.
- Happy path: discard event processed by `live-poll.mjs` updates durable session to complete.
- Happy path: a `live-poll.mjs` CLI/integration test processes an accept event and records the durable lease/result acknowledgement through the real poll script path.
- Edge case: running accept twice for the same event returns the same final durable phase without rewriting source twice.
- Error path: `live-accept.mjs` failure records an agent error in the session journal and leaves next action recoverable.
- Integration: browser reload after agent accept but before carbonize can be diagnosed by `live-status.mjs`.
- Integration: completion acknowledgement is lost after source cleanup; status detects the condition and recommends `acknowledge_completion`, then `live-complete.mjs` resolves the session complete.
**Verification:**
- Durable state reflects what happened to source, not only what browser requested.
---
- U6. **Update live-mode guidance and recovery UX**
**Goal:** Teach both agents and users the new recovery contract and make recovery visible in live mode.
**Requirements:** R2, R4, R6, R7
**Dependencies:** U3, U4, U5
**Files:**
- Modify: `skill/reference/live.md`
- Modify: `docs/adr-live-variant-mode.md`
- Modify: `skill/scripts/live-browser.js`
- Modify: `README.md` if live command usage is documented there
**Approach:**
- Update `reference/live.md` to start with a status check when resuming a session or when the agent suspects it missed events.
- Document the durable journal and session states in the ADR.
- Add concise browser states for pending accept, reconnect/recover, and agent cleanup pending.
- Keep chat overhead low: the agent should use status JSON instead of asking the user to describe browser state.
**Patterns to follow:**
- Existing `reference/live.md` contract sections for poll loop, accept, carbonize, and cleanup.
- Existing toast and bar state patterns in `skill/scripts/live-browser.js`.
**Test scenarios:**
- Test expectation: mostly documentation and UX copy. Behavioral coverage belongs to U3-U5; this unit should be verified through review plus any snapshot/E2E assertions added for visible pending states.
**Verification:**
- A new agent reading `reference/live.md` knows to recover state before making assumptions after interruption.
---
- U7. **Add restart and interruption E2E coverage**
**Goal:** Prove the durable recovery contract in realistic browser/server/agent flows.
**Requirements:** R1, R2, R3, R4, R5, R6
**Dependencies:** U1, U2, U3, U4, U5
**Files:**
- Modify: `tests/live-e2e.test.mjs`
- Modify: `tests/framework-fixtures/README.md` if new fixture expectations are needed
- Modify: `tests/live-e2e/agent.mjs` if deterministic agent hooks need to simulate interruption
**Approach:**
- Add focused E2E cases for missed polling, helper server restart, browser reload, accept before agent resumes, and carbonize-pending status.
- Prefer one or two compact fixtures over broad matrix explosion.
- Keep deterministic fake agent path as the primary verification route; LLM agent remains opt-in.
**Patterns to follow:**
- Existing live-mode E2E setup in `tests/live-e2e.test.mjs`.
- Fixture authoring guidance in `tests/framework-fixtures/README.md`.
**Test scenarios:**
- Integration: user clicks Go while no poll is active; later poll receives durable generate event and completes variants.
- Integration: user changes Tune values, reloads browser, and status/resume reports the same values.
- Integration: user clicks Accept, agent is interrupted, new agent runs status and sees `run_accept_cleanup` with variant id and params.
- Integration: helper server restarts after queued generate; resumed server can still surface the pending event.
- Integration: agent receives a poll event through the real `live-poll.mjs` path and then crashes before posting result; a resumed poll redelivers the leased event after timeout.
- Integration: source contains carbonize markers; status returns cleanup pending even without an active browser tab.
- Integration: out-of-order checkpoint after accept does not regress pending accept state.
- Error path: duplicate accept and late discard do not corrupt final source.
**Verification:**
- E2E coverage demonstrates recovery from the exact “agent was not listening” failure mode.
---
## System-Wide Impact
- **Interaction graph:** Browser `live-browser.js` posts events/checkpoints to `live-server.mjs`; server persists them through `live-session-store.mjs`; agent observes them through `live-poll.mjs`, `live-status.mjs`, or `live-resume.mjs`; source mutations still happen through `live-wrap.mjs` and `live-accept.mjs`.
- **Error propagation:** Failed event persistence must be visible to the browser before it clears state. Failed source cleanup must be visible in status output and kept recoverable.
- **State lifecycle risks:** The key risk is split-brain among browser localStorage, server journal, in-memory queue, snapshots, and source markers. The plan makes source markers plus durable journal canonical, snapshots rebuildable, in-memory queues cache-only, and browser localStorage only an optimization.
- **API surface parity:** The new status/resume scripts become agent-facing APIs and should be reflected in `reference/live.md` and tests.
- **Integration coverage:** Unit tests alone will not prove recovery. E2E must cover browser reload, server restart, missed poll, duplicate events, and carbonize pending states.
- **Unchanged invariants:** Live mode remains source-first, self-contained, zero-dependency, token-protected, and transport-compatible with current AI harnesses.
---
## Risks & Dependencies
| Risk | Mitigation |
|------|------------|
| Durable journal becomes another state source that can drift | Make source markers plus journal reconciliation explicit in `live-status.mjs`; keep browser localStorage advisory. |
| Event replay causes duplicate source rewrites | Add idempotency keys, state-machine validation, leased delivery, and tests for duplicate accept/discard. |
| Browser waits forever after accept if agent is absent | Show explicit pending/reconnect state and provide status/resume command for the agent. |
| Poll delivery is mistaken for completion | Model leased delivery separately from completion acknowledgement and redeliver expired leases. |
| Snapshot contradicts journal after crash | Treat journal as canonical and rebuild/repair snapshots on startup/status reads. |
| Annotated screenshot path goes stale before recovery | Store annotation artifacts with the incomplete session and retain them until completion cleanup. |
| Disk writes fail in restricted environments | Return structured server errors and do not clear browser state until persistence succeeds. |
| Tests become slow or brittle | Keep most coverage in store/server tests and add only targeted E2E cases for true cross-process recovery. |
| Backward compatibility with existing installed skill output drifts | Update source skill files first, then run the project build to regenerate provider outputs in a separate execution phase. |
---
## Documentation / Operational Notes
- Update `docs/adr-live-variant-mode.md` because durability changes the architecture from memory queue plus localStorage to journaled sessions.
- Update `skill/reference/live.md` so agents know to run status/resume after interruption or before assuming no pending work.
- If generated provider skill outputs are tracked, implementation should regenerate them with the existing build process after changing `skill/`.
- Because the default test script enumerates test files, implementation must update `package.json` when adding new non-E2E test files.
- Consider adding `.impeccable-live/` or the chosen session-store directory to gitignore if it is not already ignored.
---
## Sources & References
- Related architecture: `docs/adr-live-variant-mode.md`
- Live instructions: `skill/reference/live.md`
- Browser live implementation: `skill/scripts/live-browser.js`
- Server transport: `skill/scripts/live-server.mjs`
- Agent poll client: `skill/scripts/live-poll.mjs`
- Accept/discard source cleanup: `skill/scripts/live-accept.mjs`
- Live-mode tests: `tests/live-server.test.mjs`, `tests/live-accept.test.mjs`, `tests/live-e2e.test.mjs`
---
## Surgical Overhaul Follow-Ups (2026-04-29)
- Dependency audit is now Bun-native via `bun run audit`, but the gate currently fails on transitive advisories:
- `archiver -> archiver-utils -> lodash` and `brace-expansion`
- optional `puppeteer -> @puppeteer/browsers -> basic-ftp`
- Decision: keep the audit script honest and failing instead of adding ignores before exploitability and replacement cost are reviewed.
- Recommended next dependency work: evaluate whether release ZIP creation still needs `archiver`, and whether optional Puppeteer screenshot tooling can be replaced, isolated, or upgraded without bloating install size.