github-actions-advanced – skill Lập trình cho agent AI

GitHub Actions Advanced Skill

Expert guidance for designing, writing, debugging, and securing production-grade GitHub Actions workflows.

---

Detailed Guide

Read [the detailed guide](references/detailed-guide.md) before executing this skill. It retains the complete procedure and reference material. Treat its safety, prerequisites, and validation requirements as mandatory. For focused work, load the relevant sections; for end-to-end work, read the guide completely.

When to Use This Skill

• User mentions GitHub Actions, .github/workflows, CI/CD pipelines, runners, jobs, steps, or actions
• User wants to automate builds, tests, deployments, or releases via GitHub
• User asks about matrix builds, reusable workflows, composite actions, or self-hosted runners
• User needs help with OIDC authentication, caching strategies, or secrets management
• User says "my GitHub pipeline is failing" or "set up CI for my repo"
• User asks about workflow security, hardening, or environment protection rules

When NOT to Use This Skill

• The user is working with GitLab CI/CD → recommend gitlab-ci-patterns
• The user is working with CircleCI, Jenkins, or other CI platforms
• The task is purely about Docker image building without GitHub context → recommend docker-expert
• The task is about Kubernetes deployment configuration → recommend kubernetes-architect

---

Security Hardening

1. Always Declare Permissions (Least Privilege)

# Workflow-level default — restrict everything
permissions:
  contents: read

jobs:
  publish:
    # Job-level override — only expand what's needed
    permissions:
      contents: write        # Only for release/publish jobs
      packages: write        # Only for container push jobs
      pull-requests: write   # Only for PR comment jobs
      id-token: write        # Only for OIDC auth jobs

2. Pin Third-Party Actions to Full Commit SHA

# ❌ UNSAFE — tag can be mutated or hijacked
- uses: actions/checkout@v4

# Full SHA prevents tag movement; verify its origin and content (not a safety certification)
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683  # v4.2.2

# Tool to automate SHA pinning:
# npx pin-github-action .github/workflows/*.yml
# or: pip install ratchet && ratchet pin .github/workflows/

3. Prevent Script Injection

# ❌ UNSAFE — attacker controls PR title, which gets expanded in shell
- run: echo "${{ github.event.pull_request.title }}"

# ✅ SAFE — pass through environment variable (shell doesn't evaluate it)
- env:
    PR_TITLE: ${{ github.event.pull_request.title }}
  run: echo "$PR_TITLE"

# ✅ SAFE — expressions in if: conditions are evaluated by Actions, not shell
- if: github.event.pull_request.draft == false
  run: echo "Not a draft"

Never place ${{ ... }} directly inside run: when the value can come from

PR metadata, workflow inputs, repository files, matrix JSON, or earlier job

outputs. Put it in env: first, validate allowlisted values where possible, and

reference the shell variable with quotes.

4. Restrict pull_request_target Usage

# A label/association gate alone does NOT prevent untrusted execution.
# Use this event only for trusted base-code metadata tasks with narrow permissions.
on:
  pull_request_target:
    types: [labeled]

jobs:
  validate:
    # Double-guard: check label name AND author_association
    if: |
      github.event.label.name == 'safe-to-test' &&
      (github.event.pull_request.author_association == 'COLLABORATOR' ||
       github.event.pull_request.author_association == 'MEMBER' ||
       github.event.pull_request.author_association == 'OWNER')

Privileged jobs must not execute PR-controlled code or artifacts regardless of label. Check actor authorization and exact reviewed head for any separately approved action; a push after labeling changes the code. Restrict OIDC audience/subject to the exact repository and branch/environment, with minimal cloud roles. Never promote untrusted caches/artifacts into deployment code.

5. Harden with StepSecurity

# Optional third-party control: review version, permissions, supported runner and telemetry handling before adoption.
- uses: step-security/harden-runner@4d991eb9995541a0b71d1b66f1f98a5f1bef422c  # v2.11.0
  with:
    egress-policy: audit          # Start with 'audit', move to 'block' after confirming allowlist
    allowed-endpoints: >
      api.github.com:443
      registry.npmjs.org:443
      objects.githubusercontent.com:443

---

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Always test reusable workflows in a feature branch before merging to main.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.