agentleFS
nitinjain999nitinjain999/platform-skills/skills/github-actions/SKILL.md

github-actions

Skill71 starsChanged 8 days agoApache-2.0View on GitHub

Design, review, secure, and debug GitHub Actions workflows — reusable workflows, OIDC federation, SHA pinning, token scoping, promotion orchestration, and CI failure diagnosis.

  • Reads credentials
  • Commits and pushes

We split this skill into the rules it gives an agent

9 rules. 0 are common, 6 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 35 edited versions we read keep these.

  • 6

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When invoked with no arguments, ask before proceeding:

    line 27
  2. Very specific:

    Read references/github-actions.md before responding.

    line 52
  3. Distinctive:

    Every workflow should follow this job order:

    line 64
  4. Distinctive:

    Keep jobs small and named by intent.

    line 75
  5. Distinctive:

    Prefer updating Git over imperative deploys:

    line 105
  6. Distinctive:

    Add the version as a comment so reviewers can audit without resolving the SHA manually:

    line 195
  7. Very specific:

    Never print secrets: avoid echo $SECRET in run: steps

    line 228
  8. Distinctive:

    Ask the user to paste the workflow file or provide the path.

    line 239
  9. Very specific:

    Check the IAM trust policy sub condition matches the exact repo/branch/env

    line 318

“Common” is measured: kept by at least 6 in 10 of the 35 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

GitHub Actions Command

Structured guidance for designing, hardening, reviewing, and debugging GitHub Actions workflows.

Activation

/platform-skills:github-actions design    # reusable workflow or job graph design
/platform-skills:github-actions security  # OIDC, SHA pinning, token scoping, secrets hygiene
/platform-skills:github-actions review    # production-readiness checklist for an existing workflow
/platform-skills:github-actions debug     # diagnose a failing workflow or job

Interactive Wizard (fires when no mode is provided)

When invoked with no arguments, ask before proceeding:

Q1 — Mode?

What do you need?
  1. design    — reusable workflow, job graph, promotion pipeline
  2. security  — OIDC federation, SHA pinning, token scoping, secret hygiene
  3. review    — production-readiness checklist for an existing workflow file
  4. debug     — job failure, permission error, OIDC rejection, missing context

Enter 1–4 or mode name:

Q2 — Context (after mode selected):

  • design: Describe what the workflow should do — validate, build, promote, deploy?
  • security: Paste the workflow file or describe the auth pattern you are using.
  • review: Paste the workflow file or provide the path.
  • debug: Paste the error output or describe the failure symptom.

Mode: design

Triggers: design, build workflow, create pipeline, reusable workflow, job graph, promote, deploy flow

Read references/github-actions.md before responding.

Step 1 — Classify the workflow type

TypeWhen to use
Reusable workflow (workflow_call)Same job sequence needed across multiple repos or environments
Composite actionSame step sequence needed within job graphs — use /platform-skills:composite-actions
Standard workflowOne-off or repo-specific, not shared

Step 2 — Apply the canonical job pattern

Every workflow should follow this job order:

validate → build → [test] → promote → [deploy]
  • validate: lint, format, policy gates, unit checks — fast, no credentials
  • build: image or artifact packaging — OIDC to registry
  • promote: update version pin or overlay in Git — triggers GitOps reconciler
  • deploy: guarded apply only if not using GitOps

Keep jobs small and named by intent. If a job is hard to name, it is doing too much.

Step 3 — Reusable workflow structure

# .github/workflows/validate.yml — called by other workflows
on:
  workflow_call:
    inputs:
      environment:
        required: true
        type: string
        description: "Target environment (dev | staging | prod)"
    secrets:
      token:
        required: true

jobs:
  validate:
    runs-on: ubuntu-latest
    permissions:
      contents: read
    steps:
      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7.0.0
        with:
          token: ${{ secrets.token }}

Step 4 — Promotion orchestration

Prefer updating Git over imperative deploys:

# promote job — updates the version pin in the GitOps repo
promote:
  needs: build
  runs-on: ubuntu-latest
  permissions:
    contents: write
  steps:
    - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7.0.0
    - name: Update image tag
      env:
        NEW_TAG: ${{ needs.build.outputs.image_tag }}
      run: |
        sed -i "s|image:.*|image: ghcr.io/org/app:${NEW_TAG}|" \
          deploy/overlays/${{ inputs.environment }}/kustomization.yaml
        git config user.email "ci@org.com"
        git config user.name "CI"
        git commit -am "chore: promote app to ${NEW_TAG} in ${{ inputs.environment }}"
        git push

Handoffs:

  • Extracting repeated steps → /platform-skills:composite-actions
  • Terraform plan/apply in the workflow → /platform-skills:terraform
  • GitOps reconciler that picks up the promotion commit → /platform-skills:gitops

Mode: security

Triggers: OIDC, pin, SHA, token, permissions, secrets, secure, harden, federation

Read references/github-actions.md before responding.

Repositories with Kubernetes manifests or Helm charts: the posture gate is /platform-skills:kubescape ci (signature-verified CLI install, fails PRs only on new findings). Do not use kubescape/github-action: its latest release bundles CLI v3.0.21 and its README pins @main.

OIDC federation — AWS

permissions:
  id-token: write    # required for OIDC token request
  contents: read

steps:
  - uses: aws-actions/configure-aws-credentials@254c19bd240aabef8777f48595e9d2d7b972184b  # v6.2.1
    with:
      role-to-assume: arn:aws:iam::123456789012:role/github-actions-deploy
      aws-region: eu-north-1

IAM trust policy (scope to repo and branch):

{
  "Condition": {
    "StringLike": {
      "token.actions.githubusercontent.com:sub": "repo:org/repo:ref:refs/heads/main"
    }
  }
}

OIDC federation — Azure

permissions:
  id-token: write
  contents: read

steps:
  - uses: azure/login@532459ea530d8321f2fb9bb10d1e0bcf23869a43  # v3.0.0
    with:
      client-id: ${{ secrets.AZURE_CLIENT_ID }}
      tenant-id: ${{ secrets.AZURE_TENANT_ID }}
      subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }}

SHA pinning checklist

# Resolve a tag to its current commit SHA
gh api repos/{owner}/{repo}/git/refs/tags/{tag} \
  --jq '.object.sha'

# For actions that use a commit SHA directly (not a tag ref):
gh api repos/{owner}/{repo}/commits/{tag} --jq '.sha'

Add the version as a comment so reviewers can audit without resolving the SHA manually:

uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7.0.0

Token scoping — minimum permissions per job

Job typePermissions needed
Read-only checkoutcontents: read
Push a commitcontents: write
Create or comment on PRpull-requests: write
Publish packagespackages: write
OIDC token requestid-token: write
Read Issuesissues: read

Set permissions: {} at the workflow level to deny all, then grant per-job:

permissions: {}   # deny all at workflow level

jobs:
  build:
    permissions:
      contents: read
      id-token: write
      packages: write

Secret hygiene

  • Store secrets in GitHub Environments, not at repository level, for deployment secrets
  • Never print secrets: avoid echo $SECRET in run: steps
  • Mask dynamic secrets: echo "::add-mask::${DYNAMIC_SECRET}"
  • Rotate long-lived tokens on a schedule; prefer OIDC to eliminate them
  • Audit secret usage: gh secret list and gh secret list --env <env>

Mode: review

Triggers: review, audit, checklist, is this workflow safe, production ready

Ask the user to paste the workflow file or provide the path. Then evaluate against:

CRITICAL (must fix before merge)

❌ External `uses:` with mutable tag (@v4, @main, @latest) — supply chain risk; pin to full SHA
❌ `permissions: write-all` or no permissions block — grants all scopes by default
❌ `${{ github.event.*.body }}` or PR title interpolated in run: — script injection risk
❌ Long-lived cloud credentials stored as secrets — replace with OIDC
❌ `pull_request_target` with checkout of the PR head — arbitrary code execution risk

WARNING (should fix)

⚠️  No concurrency group — parallel runs on same branch can conflict or double-deploy
⚠️  No timeout-minutes on jobs — a hung job blocks runners indefinitely
⚠️  Secrets passed to untrusted third-party actions — scope secrets to first-party or audited actions only
⚠️  Environment protection rules not used for production deploys
⚠️  No required status checks on the branch protection rule this workflow targets

INFORMATIONAL

ℹ️  No dependabot.yml for github-actions ecosystem — action versions will drift
ℹ️  Reusable workflow could replace copy-pasted job sequence
ℹ️  Job name does not describe intent clearly

Report format:

## Review: <workflow-name>

### CRITICAL (N findings)
❌ Line 12: uses: actions/checkout@v4 — pin to SHA. Current SHA: <sha>

### WARNING (N findings)
⚠️  No concurrency group defined

### INFORMATIONAL (N findings)
ℹ️  No dependabot.yml for the github-actions ecosystem

### Score: N/15 — Poor | Fair | Good | Excellent

Mode: debug

Triggers: failing, error, permission denied, OIDC, 401, 403, missing context, debug, not working

Identify the failure layer first:

1. Syntax / parse error          → workflow YAML invalid
2. Permission / auth error       → token scoping, OIDC trust, secret missing
3. Missing context               → expression evaluates to empty, wrong event trigger
4. Action version mismatch       → SHA pinned to a version with a breaking change
5. Runner environment            → tool not available on runner image
6. Downstream service error      → cloud API, registry, or deploy target failing

Diagnosis by error type

Resource not accessible by integration / 403

# Check the permissions block — add the missing scope
permissions:
  contents: write       # missing if you get 403 on git push
  pull-requests: write  # missing if you get 403 on PR comment
  id-token: write       # missing if OIDC token request fails

OIDC rejection (Not authorized to perform sts:AssumeRoleWithWebIdentity)

  1. Confirm id-token: write is set on the job
  2. Check the IAM trust policy sub condition matches the exact repo/branch/env
  3. Confirm the workflow is running from the branch/environment specified in the condition
  4. For environment-scoped trust: repo:org/repo:environment:production — job must use environment: production
# Decode the OIDC token to inspect claims (add a debug step temporarily)
- name: Inspect OIDC token claims
  run: |
    TOKEN=$(curl -sSfL -H "Authorization: bearer $ACTIONS_ID_TOKEN_REQUEST_TOKEN" \
      "${ACTIONS_ID_TOKEN_REQUEST_URL}&audience=sts.amazonaws.com" | jq -r '.value')
    echo "${TOKEN}" | cut -d. -f2 | base64 -d 2>/dev/null | jq

Expression evaluates to empty

# Workflow inputs are empty in non-workflow_call events
# Guard with:
- if: ${{ inputs.environment != '' }}

Context access might be invalid

Caused by accessing a context that does not exist for the triggering event (e.g. github.event.pull_request on a push event). Check which contexts are available for the trigger.

Validation command

# Lint the workflow file locally before pushing — syntax, expressions, shell
actionlint .github/workflows/<workflow>.yml

# Audit the same file for security findings — a valid workflow can still be unsafe
zizmor --strict-collection .github/workflows/<workflow>.yml

# Check recent run failures with full logs
gh run list --workflow <workflow>.yml --limit 10
gh run view <run-id> --log-failed

actionlint and zizmor do not overlap. actionlint answers "is this workflow valid" (syntax, expression types, shellcheck on run:). zizmor answers "is this workflow safe" (template injection, credential persistence, unpinned uses:, over-broad permissions:, impostor commits). Run both. For the full audit catalogue, policy file, and CI gate: /platform-skills:zizmor.


Common mistakes

  • pull_request_target + PR head checkout — runs with write permissions and access to secrets; combined with checkout of untrusted code this allows secret exfiltration. Never checkout ${{ github.event.pull_request.head.sha }} in a pull_request_target workflow.
  • Mutable action tags — @v4 can be rewritten by the action author; a supply chain compromise silently executes attacker code. Always pin to SHA.
  • Over-scoped GITHUB_TOKEN — default permissions can be broad depending on repo settings. Always set an explicit permissions: block.
  • Forgetting OIDC requires the job's environment to match the trust policy — if the trust uses environment:production, the job must declare environment: production or the token request is rejected.
  • workflow_dispatch with no input validation — inputs are free-text by default. Use type: choice for enum inputs and validate string inputs in the first step.

Reference

Full guidance: references/github-actions.md

For composite action scaffolding and review: /platform-skills:composite-actions

For workflow security auditing with zizmor (audit catalogue, zizmor.yml policy, auto-fix, CI gate, pre-commit): /platform-skills:zizmor

Examples:

  • examples/github-actions/terraform-cicd.yml — Terraform plan + apply with OIDC
  • examples/github-actions/container-build.yml — Docker build + GHCR push
  • examples/github-actions/reusable-workflows/ — reusable workflow patterns

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.