--- name: openiap-workflows description: Use for OpenIAP monorepo work that should follow the repository's shared agent workflows, including review-pr, audit-code, audit-security, audit-iapkit, compile-knowledge, verify-all, e2e-tests, stable or prerelease package releases, resolve-issue, commit/push/PR, generated type sync, package-specific checks, GitHub review threads, and project conventions from AGENTS.md. --- # OpenIAP Workflows Use this skill when the user asks the agent to perform an OpenIAP repo workflow that previously lived under `.claude/commands`, such as reviewing a PR, resolving an issue, auditing code, compiling knowledge, running device-backed E2E regression, verifying the monorepo, or committing and opening a PR. ## Source Of Truth Before changing code, read the root `AGENTS.md`; `CLAUDE.md` and `GEMINI.md` are symlinks to it in this repo, while Grok, Codex, and Muse consume `AGENTS.md` directly. Then read the relevant detailed files: - Package and library rules: `knowledge/internal/*.md` - Package conventions: `packages/*/CONVENTION.md` - Library conventions: `libraries/*/AGENTS.md` - Workflow details: `.claude/commands/*.md` Do not duplicate or reinterpret those rules when a file already covers the specific package or workflow. ## Command Mapping You do not need Claude slash-command syntax. If the user says any of these natural-language requests, execute the matching workflow: - Review PR comments, fix review feedback, or "review-pr": read `.claude/commands/review-pr.md`. CodeRabbit is the only external reviewer that posts to the PR; do not invoke other review bots. If CodeRabbit cannot review the current head, run the single-round Codex fallback that command defines; read `.codex/skills/review-self/SKILL.md` and run its single-round `review-pr` fallback only when Codex is unavailable too. At the clean end of the loop, remove the temporary CodeRabbit trigger and terminal skip/unavailable top-level comments exactly as defined by the command workflow. - Audit code, check latest APIs, or "audit-code": read `.claude/commands/audit-code.md`. - Audit SBOM quality, release provenance, workflow permissions, or supply-chain/security posture: read `.claude/commands/audit-security.md`. - Reconcile the IAPKit site with OpenIAP, audit kit docs, or check whether IAPKit reflects a spec/store update: read `.claude/commands/audit-iapkit.md`. - Compile knowledge or rebuild AI context: read `.claude/commands/compile-knowledge.md`. - Resolve a GitHub issue: read `.claude/commands/resolve-issue.md`. - Verify all, health check, or pre-PR verification: read `.claude/commands/verify-all.md`. - E2E tests, device regression, connected-device purchase flow checks, or "e2e-tests": read `.claude/commands/e2e-tests.md`. - Android E2E tests, Play/Amazon/Horizon/VegaOS regression, or "e2e-tests-google": read `.claude/commands/e2e-tests-google.md`. - Apple E2E tests, iOS regression, or "e2e-tests-apple": read `.claude/commands/e2e-tests-apple.md`. - Stable releases, RC/next releases, registry publication, or package deploys: read `.claude/commands/release.md`. - Commit, push, or create PR: read `.claude/commands/commit.md`. When a command file gives a sequence, follow it unless the user's newest instruction narrows the scope. For `e2e-tests`, an unqualified request means the full regression matrix in the command file, including native packages, framework libraries, build-only platform rows, connected-device rows, and explicit blocked/unsupported rows. A request naming one platform runs only that half's scoped file. ## Internal Workflow Change Guard Internal agent/workflow-only changes include `.claude/commands/`, `.claude/skills/`, `.codex/skills/`, `.cursor/rules/`, `AGENTS.md`, `CLAUDE.md`, `GEMINI.md`, and agent automation notes. Do not create a branch, push, or open a PR for those changes unless the user explicitly asks to publish, PR, or merge them. If a user asks to update an internal workflow and does not explicitly ask for a PR, keep the change local and report the changed files. If a PR is already open for an internal workflow because the user explicitly requested it, add appropriate labels before merging. ## Non-Negotiables - Apply the canonical KISS/SSOT release criteria in `knowledge/internal/03-coding-style.md`. - Before any public GitHub write, apply the English-only communication guard in `knowledge/internal/06-git-deployment.md`. Private maintainer conversation language must never leak into issue, PR, review, release, or commit prose. - Read relevant knowledge and package convention files before editing package or library code. - Never hand-edit generated files unless the workflow explicitly says to verify generated output after running the generator. - For GraphQL schema/API changes, follow the SDK Parity Checklist in `knowledge/internal/04-platform-packages.md`. - Run the package-specific verification commands for touched paths. - For dependency modernization release trains, keep the entire train in one PR and follow `.claude/commands/release.md` for the all-workflow preflight. If a post-merge stable release reveals a CI-only blocker, pause the train, inspect all remaining workflows, and prepare the confirmed repairs together. Apply `knowledge/internal/06-git-deployment.md#opening-pull-requests` before opening a recovery PR; release authorization alone does not authorize that new PR. - For Android package work, compile Play, Horizon, and Amazon variants when relevant. - For docs/API/type docs changes, run `bun audit:docs` or the documented audit command before pushing. - For release-note package lists, verify versions from package metadata and GitHub release tags; never infer framework versions from `openiap-versions.json` or from a nearby release block. - Before creating or updating a PR, declaring review clean, or releasing, apply `knowledge/internal/05-docs-patterns.md#release-note-completeness-gate`. - Run stable and prerelease package releases from `main` through an explicit dispatch. RCs use npm `next`; stable uses `latest`. Vercel automatically deploys main's docs, including RC metadata. Run `bun run audit:release-state` before release work. - For PRs with new features, visible behavior changes, UI changes, docs pages, example flows, or developer workflows, record the actual changed surface, compress the video to under 10 MB, and upload it to the GitHub PR as a `Preview` comment or PR body attachment. Never commit one-off preview media, including under `.github/pr-previews/`; keep it in a temporary or ignored local path and delete it after verifying the GitHub attachment. If browser or extension permissions block the attachment, stop and ask the maintainer to enable uploads instead of force-adding a Git fallback. Use the Codex Chrome Extension for web/docs/dashboard previews when applicable. - Keep commits in Angular Conventional Commits format: `(): `. - When creating branches for commit/push/PR workflows, follow `.claude/commands/commit.md`: use meaningful semantic prefixes such as `feat/`, `fix/`, `ci/`, `docs/`, `test/`, `chore/`, or `refactor/`. Do not use generic agent/tool prefixes such as `codex/`. ## GitHub Review Threads For PR review feedback, use the GitHub app tools or `gh` as needed to inspect inline review threads. Fix valid findings in the current PR, reply to the specific inline comment with the plain commit hash, and resolve only threads that are fixed or outdated per `.claude/commands/review-pr.md`. Do not call a PR clean merely because CodeRabbit skipped or failed. Do not replace it with another external review bot. Use the head-specific Codex fallback required by the command workflow — or the `review-self` fallback when Codex is unavailable too — and include its clean result in the completion gate. Do not reply with "will address later" for valid correctness or operational findings. Implement the fix in the current PR unless the finding is wrong on the merits, and explain the concrete repo evidence when pushing back.