## Summary
Mirrors [openai/codex#28389](https://github.com/openai/codex/pull/28389)
into the system and curated OpenAI Docs skills. The compact search
guidance performs better in Codex.
## Validation
- Both skill validations passed
- `git diff --check`
## Summary
- mirror the OpenAI docs skill changes from
[openai/codex#24914](https://github.com/openai/codex/pull/24914) into
both the system and curated `openai-docs` copies
- add Codex self-knowledge routing, manual-first lookup guidance, and
updated skill metadata
- add the cached Codex manual fetch helper with temp-cache
auto-discovery and hash verification
## Why
The standalone skills repository should ship the same Codex-aware OpenAI
docs behavior already merged into the Codex bundled skill while
preserving the existing `.system` and `.curated` distinction.
## Impact
OpenAI docs lookups can now route broad Codex questions through the
current Codex manual before using Docs MCP fallback. The helper works
without requiring an explicit `--cache-dir` override.
## Validation
- `git diff --check`
- `node --check
skills/.curated/openai-docs/scripts/fetch-codex-manual.mjs`
- `node --check
skills/.system/openai-docs/scripts/fetch-codex-manual.mjs`
- live `curl -fsSI https://developers.openai.com/codex/codex-manual.md`
verification returned `x-content-sha256`
- helper smoke test without `--cache-dir`: fresh fetch returned
`cacheStatus: "updated"`, second run returned `cacheStatus: "hit"`
- exact helper and agent metadata comparison across `.system` and
`.curated` copies
## Summary
- Add the `define-goal` curated skill for shaping user intent into
concrete, measurable goals.
- Include agent UI metadata so the skill can be invoked as
`$define-goal`.
- Add the standard Apache 2.0 license file for the new skill package.
## Validation
- `quick_validate.py
/Users/richardlee/code/skills/skills/.curated/define-goal`
## Why
The curated `yeet` skill was missing guidance for common publish-flow
cases: repositories with PR templates, branches that already have a PR,
and PR bodies that need to preserve reviewer-relevant context instead of
generic boilerplate.
## What Changed
Updated `skills/.curated/yeet/SKILL.md` to make the publish workflow
template-aware and branch-aware. The skill now tells Codex to discover
the active repository PR template before creating a PR, reuse an
existing PR for the current branch instead of creating a duplicate,
preserve an existing PR's draft or ready-for-review state, apply
conventional PR title guidance, and write PR bodies around the net
change with meaningful verification only when there is behavioral
evidence worth preserving.
## Summary
- Add OpenAI API key setup coordination guidance to the public
openai-docs skill
- Prefer the openai-platform-api-key skill before docs/API example work
for API-backed implementation tasks
- Keep docs-only guidance routed directly to openai-docs
## Validation
- git diff --check origin/main..HEAD
---------
Co-authored-by: Codex <noreply@openai.com>
## Summary
- Improve hatch-pet discovery for plural searches by adding the literal
word `pets` to the recommended-skill description
- Update the agent short description with the same plural wording so
both metadata surfaces match
## Discovery context
- Codex App recommended skill search is substring-based over `name`,
`description`, and `shortDescription`; it does not stem `pet` to `pets`
- `hatch-pet` previously had `pet spritesheets`, which matches `pet` but
not contiguous `pets`
## Validation
- `rg -n "description:.*pets|short_description:.*pets|animated pets"
skills/.curated/hatch-pet/SKILL.md
skills/.curated/hatch-pet/agents/openai.yaml`
## Summary
- Clarify that the non-directional hatch-pet `running` row is an active
working/in-progress task loop, not literal foot-running
- Keep `running-right` and `running-left` as the only directional
locomotion rows
- Update the generated row prompt requirements so image generation
rejects jogging, sprinting, treadmill poses, raised knees, long steps,
pumping arms, and travel for the `running` row
- Port the local low-distraction `idle` guidance into hatch-pet without
replacing Gabriel's subagent/imagegen workflow: idle now asks for subtle
breathing/blinking/bobbing only and rejects distracting action states
## Validation
- `python3 -m py_compile
skills/.curated/hatch-pet/scripts/prepare_pet_run.py`
- `python3 skills/.curated/hatch-pet/scripts/prepare_pet_run.py
--pet-name Testy --description "A tiny test pet." --pet-notes "a tiny
blue test pet" --output-dir
/private/tmp/openai-skills-hatch-running-prompt-check --force`
- `rg -n "active working|literal foot-running|running a task|generic
in-place|Show in-place running|jogging|sprinting"
/private/tmp/openai-skills-hatch-running-prompt-check/prompts/rows/running.md
/private/tmp/openai-skills-hatch-running-prompt-check/imagegen-jobs.json`
- `python3 skills/.curated/hatch-pet/scripts/prepare_pet_run.py
--pet-name Testy --description "A tiny test pet." --pet-notes "a tiny
blue test pet" --output-dir
/private/tmp/openai-skills-hatch-idle-prompt-check --force`
- `rg -n "low-distraction|subtle idle|same pose|waving, walking|first
and last|active working|literal foot-running"
/private/tmp/openai-skills-hatch-idle-prompt-check/prompts/rows/idle.md
/private/tmp/openai-skills-hatch-idle-prompt-check/prompts/rows/running.md`
**Summary**
- Clarify that the `migrate-to-codex` skill should finish by marking
completed TODO steps as `completed`
- Explicitly prevent a completed migration from leaving a stale
`in_progress` task behind
### Motivation
- Remove the curated Sora skill package and all associated runtime/docs
so it no longer appears in the skills catalog.
### Description
- Delete the entire `skills/.curated/sora` directory, removing
`SKILL.md`, `agents/openai.yaml`, `scripts/sora.py`, `assets/*`,
`references/*`, and `LICENSE.txt`.
- Remove the bundled CLI, reference docs, assets, and agent manifest so
the skill is no longer discoverable or installable from this repository.
- No other code paths or files required modifications because skill
discovery is directory-based.
### Testing
- Ran `rg -n "\bsora\b|Sora" skills README.md contributing.md` and
confirmed there are no remaining references (success).
- Verified the directory removal with `test ! -d skills/.curated/sora`
which reported the skill directory is gone (success).
- Checked the repository status with `git status --short` to ensure the
deletions are reflected in the working tree (success).
------
[Codex
Task](https://chatgpt.com/codex/cloud/tasks/task_i_69f193d7125c833196851fb3b90fc534)
**Summary**
- Adds the public `migrate-to-codex` curated skill for migrating
external agent setup into Codex-facing files.
- Includes the approved skill instructions, references, and migration
scripts from openai/skills-internal#84.
- Places the skill at `skills/.curated/migrate-to-codex`, which is the
installable public path the Codex App should reference.
## Summary
- Clarifies reasoning-effort recommendation wording
- Replaces internal snake_case prompt block names with natural-language
guidance aligned to the prompting guide
## Test plan
- `git diff --check`
- Verified the old snake_case prompt block names no longer appear in the
upgrade guide
## Summary
This basically copies the OpenAI docs skill from `skills/.system` to
`skills/.curated` since this wasn't in my previous PR merging changes to
`skills/.curated`.
This copies the GPT-5.4 freshness-check structure from
`skills/.system/openai-docs` into `skills/.curated/openai-docs`,
including the model-agnostic reference filenames and resolver script.
## Test Plan
- Verified `skills/.curated/openai-docs` and
`skills/.system/openai-docs` are identical with `diff -rq`.
- Ran `node --check
skills/.curated/openai-docs/scripts/resolve-latest-model-info.js`.
- Verified the resolver parses a local GPT-5.4 `latestModelInfo`
fixture.
- Scanned the diff for unrelated launch/model language.
## Summary
- Add a helper script that reads latest model metadata from
developers.openai.com
- Check that metadata before applying the bundled GPT-5.4 upgrade guide
- Refresh the bundled GPT-5.4 prompting guide snapshot from the public
Prompt guidance page
- Fall back to bundled GPT-5.4 references when remote metadata or guides
are unavailable
## Test plan
- node --check
skills/.system/openai-docs/scripts/resolve-latest-model-info.js
- Tested the resolver against local GPT-5.4 latest-model metadata
- git diff --check
## Summary
- Move the `imagegen` skill from `skills/.curated` into `skills/.system`
- Preserve the skill contents and assets under the new installed
location
- Update the CLI reference to point at the new system path
## Testing
- Not run (not requested)
**Summary**
- Adds curated `$cli-creator` skill for building composable CLIs Codex
can run from any repo.
- Defines the command contract: PATH install, `doctor`, bounded JSON,
discovery/resolve/read/write commands, raw request fallback, and
auth/config handling.
- Adds runtime guidance, reference patterns, and the companion-skill
handoff with `$skill-creator`.
Updated the `yeet` skill so the GitHub artifacts it creates use neutral
naming instead of assistant-branded naming.
The skill now recommends `{description}` for new branches and uses the
plain `{description}` as the PR title. The workflow example was updated
to match the new branch naming convention.
## Summary
- Replace the custom `sentry_api.py` script with native Sentry CLI
commands, which handle authentication, org/project auto-detection,
pagination, and retries out of the box.
- Update `SKILL.md` with CLI-based workflows covering issue inspection,
trace exploration, log streaming, and a generic API fallback for
uncovered endpoints.
- Update the agent description in `openai.yaml` to reflect the CLI-based
approach.
## Motivation
The bundled Python script was a thin wrapper around Sentry's REST API
that required manual auth token setup and explicit org/project
configuration. The Sentry CLI (`sentry-cli`) handles all of this
natively — including interactive auth, DSN-based project detection, and
structured JSON output — making the script redundant and the skill
simpler to use.
## What changed
- replace the bundled Apache license text in `skills/.curated/figma` and
`skills/.curated/figma-implement-design` with the Figma Developer Terms
notice
- add the same Figma license notice to the other curated Figma skill
directories that were missing a per-skill license file
## Why
These curated Figma partner skills need to ship with the correct
Figma-specific licensing notice in each skill directory instead of a
generic Apache license or no per-skill license file.
## Impact
Consumers of the curated Figma skills now see the correct terms directly
in each affected package.
## Validation
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/figma`
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/figma-implement-design`
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/figma-use`
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/figma-code-connect-components`
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/figma-create-design-system-rules`
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/figma-create-new-file`
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/figma-generate-design`
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/figma-generate-library`
Co-authored-by: Ashwin Mathews <ashwinm@openai.com>
Document how ChatGPT widgets should handle uploads, library reuse, and
file download URLs.
## Current state
The ChatGPT apps reference lists the core `window.openai` helpers, but
it does not explain the newer file-library flows or when app authors
should choose one file API over another.
That leaves widget implementations to guess between local upload,
library-backed upload, and existing-file selection.
## What this PR does
- updates the API surface list to show `uploadFile(file, options?)`
- documents `selectFiles()` alongside the other ChatGPT-only helpers
- adds a file-patterns section describing when to use plain uploads,
library saves, library selection, and `getFileDownloadUrl`
## Riskiest parts
This is a docs-only change, so the main risk is semantic accuracy if the
described file-helper behavior drifts from the host implementation.
## Testing
- Searched the repo for a checked-in automated test or CI entrypoint and
did not find one: no root `package.json`, `pyproject.toml`, `Makefile`,
`justfile`, workflow files, or documented test commands.
- No automated tests were run because this repo does not appear to
expose an automated test harness.
## Manual testing
- Review the rendered markdown for
`skills/.curated/chatgpt-apps/references/window-openai-patterns.md`.
- Confirm the guidance matches current ChatGPT widget file flows for
upload, library save, selection, and download URL access.
## Spiciness
1/5
Co-authored-by: Codex <noreply@openai.com>
## What changed
- add six curated Figma skills: `figma-use`, `figma-generate-design`,
`figma-generate-library`, `figma-code-connect-components`,
`figma-create-design-system-rules`, and `figma-create-new-file`
- refresh the existing `figma` and `figma-implement-design` skill
metadata to match the imported partner workflow split
- import the referenced assets, maintainers files, docs, and helper
scripts that those skills depend on
- add a small follow-up cleanup commit to remove repo-unsupported
`disable-model-invocation` frontmatter and strip imported trailing
whitespace
## Why
- Figma asked for the latest skills from `figma/mcp-server-guide` to be
available in `openai/skills`
- the provided patch needed a small repo-specific cleanup pass to
satisfy this repository's skill validation rules
## Impact
- makes the expanded Figma skill set available from the curated skills
catalog
- keeps the imported skills structurally compatible with the current
`openai/skills` validation expectations
## Validation
- `python3 skills/.system/skill-creator/scripts/quick_validate.py` for
`skills/.curated/figma-code-connect-components`,
`skills/.curated/figma-create-design-system-rules`,
`skills/.curated/figma-create-new-file`,
`skills/.curated/figma-generate-design`,
`skills/.curated/figma-generate-library`, `skills/.curated/figma-use`,
and `skills/.curated/figma-implement-design`
- `git diff --check origin/main...HEAD`
- `python3 -m py_compile
skills/.curated/figma-code-connect-components/scripts/normalize_node_id.py`
- `bash -n
skills/.curated/figma-create-design-system-rules/scripts/check_agents_md.sh`
- `node --check skills/.curated/figma-generate-library/scripts/*.js`
---------
Co-authored-by: Ashwin Mathews <ashwinm@openai.com>
## Summary
- add a new curated `frontend-skill` derived from the GPT-5.4 frontend
design guidance
- package the skill as a minimal curated bundle with `SKILL.md`,
`agents/openai.yaml`, and `LICENSE.txt`
- keep the skill focused on composition, hierarchy, imagery, copy, and
motion guidance for visually strong front-end work
## Why
Skill provides Codex stronger guidance on art direction, structure, and
interaction quality.
## Validation
- `python3 skills/.system/skill-creator/scripts/quick_validate.py
skills/.curated/frontend-skill`
- Parsed `skills/.curated/frontend-skill/agents/openai.yaml` with
`yaml.safe_load(...)`
## Summary
- Refreshes the Sora skill to match the latest Sora docs and changelog,
including support guidance for non-human reusable characters, `16`/`20`
second clips, `1920x1080` and `1080x1920` on `sora-2-pro`, extensions,
JSON/file image references, and clearer local queue vs official Batch
API guidance.
- Simplifies the supported skill surface so it only advertises broadly
available workflows: create videos, create non-human character
references, extend completed videos, edit existing generated videos by
ID, and manage/download/list/delete jobs.
- Removes eligibility-gated paths and messaging from the skill:
uploaded-video editing is no longer exposed in the CLI or docs, and the
docs no longer mention account-manager/sales escalation or
human-likeness eligibility.
## Implementation details
- Updated `skills/.curated/sora/scripts/sora.py` to add the new Sora
flows and limits, while narrowing `edit` to `--id` only and rejecting
the old `--video-file` path.
- Updated `skills/.curated/sora/SKILL.md` plus the
CLI/API/troubleshooting references to reflect the current supported
workflow and non-human character boundary.
- Regenerated the Sora agent metadata so the UI description/prompt
matches the expanded but non-gated skill surface.
## Summary
- align the `.system` and `.curated` `latest-model.md` references
- update the text model rows to use `gpt-5.4-mini` and `gpt-5.4-nano`
- clarify that `gpt-5.4` and `gpt-5.4-mini` are suitable for coding
use-cases
- remove outdated no-reasoning and Codex-specific rows from the table
## Testing
- Not run (not requested)
Ensure that as https://github.com/openai/codex/pull/14175 rolls out,
this skill will take advantage of it (adding the arg before the change
rolls out is a no-op, so this is safe to add now).
Broaden the skill docs to cover reusable web, mobile, and Electron
session helpers, clearer guidance on explicit viewport versus
native-window passes, and safer reload and restart patterns.
Add model-bound screenshot normalization guidance for web and Electron,
including CSS-coordinate click and tap flows plus Electron-specific
capture caveats.
Co-authored-by: Codex <noreply@openai.com>
## Summary
- add a dedicated small SVG icon for the curated `slides` skill
- add a 100x100 PNG large icon asset for the skill card
- wire both assets into `skills/.curated/slides/agents/openai.yaml` so
the skill uses the new branding in the UI
## Validation
- `git diff --check`
- verified `skills/.curated/slides/assets/slides-small.svg` is valid SVG
- verified `skills/.curated/slides/assets/slides.png` is a 100x100 PNG
Co-authored-by: Ed Bayes <@openai.com>
## Summary
- update the existing curated OpenAI Docs skill to match the new
GPT-5.4-aware guidance
- add bundled reference docs for model selection and GPT-5.4
migration/prompt-upgrade flows
- refresh the skill metadata and workflow text to load those references
when relevant
## Testing
- not run (skill content update only)
## Summary
- add the OpenAI Docs skill under skills/.system/openai-docs
- include the skill metadata, assets, and GPT-5.4 upgrade reference
files
- exclude the test harness and test fixtures
## Testing
- not run (skill-only asset copy)
## Summary
- add a new curated `chatgpt-apps` skill for building, scaffolding,
refactoring, and troubleshooting ChatGPT Apps SDK apps
- define a docs-first workflow that pulls current OpenAI Apps SDK docs
before generating code
- add reference guides for app archetypes, upstream example selection,
`search`/`fetch` conventions, MCP Apps bridge and `window.openai`
patterns, and repo validation expectations
- include a low-dependency scaffold script that generates a minimal Node
+ `@modelcontextprotocol/ext-apps` starter with an MCP server and widget
UI
- bundle agent metadata and license text for the new curated skill
## Why
This gives Codex a reusable, opinionated workflow for ChatGPT app work
instead of ad hoc scaffolding. It pushes the model toward current docs,
the smallest matching upstream example, and a stable runnable repo
contract.
Summary
- document a preliminary `vercel` CLI check and clarify when to escalate
permissions
- align deploy script references with other skills using the `skill_dir`
placeholder for script paths
- spell out fallback behavior when the CLI is missing or lacks
credentials
Testing
- Not run (not requested)
## Summary
- add `skills/.experimental/wrapped`
- include skill definition, agent config, and scripts for wrapped
reports
## Testing
- not run (file import only)
Co-authored-by: Ed Bayes <@openai.com>