GitHub Actions
Your agent now writes and debugs GitHub Actions workflows against the official GitHub documentation — syntax, triggers, matrices, runners, reusable workflows, caching, secrets, OIDC, deployments, and custom actions.
What it installs
| Artifact | Path (in your project) | Purpose |
|---|---|---|
| Skill | .agents/skills/github-actions-docs/ | Protocol the agent follows on Actions work — grounds answers in official GitHub docs instead of stale memory. Sourced from xixu-me/skills |
| Rule | .claude/rules/github-actions.md | Always-loaded pointer: triggers on any workflow file, Actions YAML, or CI/CD question |
How it works
GitHub Actions has a deceptively large surface — trigger semantics, matrix combinatorics, artifact/cache scope rules, OIDC federation for cloud deploys, reusable vs. composite actions, and permissions models change often enough that LLM memory drifts. The skill's job is to ground answers: fetch the right GitHub docs page and return docs-backed YAML rather than plausible-sounding guesses.
The rule loads a pointer into every session, so the agent consults the skill before writing on: triggers, matrix strategies, permissions: blocks, uses: references, or deployment jobs — no explicit invocation needed.
Setup
No env vars or external accounts required. The skill is fetched from GitHub during harness-kit add via npx skills add.
Pairs well with
pre-commit-hooks— run workflow linters (actionlint) locally before pushing CI changescommit-signing— OIDC-authenticated signing fits naturally with Actions deploy jobscode-review-gates— enforce secret/permissions checks on every workflow diff