DEV Community

2B0748
2B0748

Posted on

I Built an AI Skill That Writes GitHub Actions Workflows — With Zero Config Errors

#github #ai #automation #devops

I Built an AI Skill That Writes GitHub Actions Workflows — With Zero Config Errors

Stop hand-crafting YAML. Let any AI agent generate production-ready workflows with 12 quality gates, 5 decision templates, and 3 mandatory security configs.

The Problem

Hand-writing GitHub Actions YAML is error-prone. Did you remember permissions? Did you pin action versions? Did you configure caching? Is concurrency set?

Every omitted config is either slower builds, security holes, or wasted billable minutes.

The Solution

I built an AI Skill — an executable instruction set, not a vague prompt. It forces any AI agent to follow a strict pipeline:

Decision Tree → Mandatory Config Injection → Cache Matching → 12 Quality Gates → Production YAML
Enter fullscreen mode Exit fullscreen mode

One sentence, instant result:

"Add CI for my Node.js project — run eslint and jest on PRs"

What the AI generates under the hood:

  1. ✅ Detects package.json → Node.js
  2. ✅ Template A (CI) selected
  3. ✅ permissions: contents: read injected
  4. ✅ concurrency with auto-cancel
  5. ✅ cache: 'npm' → ~70% faster installs
  6. ✅ All actions SHA-1 pinned (@11bd71... not @v4)
  7. ✅ 12-point checklist validated

5 seconds. Zero omissions.

What Sets It Apart

Approach Universal Security Check Caching Quality Gates
GitHub templates ✅ ❌ ❌ ❌
Manual YAML ✅ ❌ ❌ ❌
Raw ChatGPT prompt ✅ ❌ ❌ ❌
This Skill ✅ All agents ✅ 5 red lines ✅ 6 langs ✅ 12 gates

Raw LLM chats miss configs every time. This Skill makes them mandatory, not optional.

5 Decision Paths Cover 95% of Use Cases

  • "lint / test / PR" → CI Workflow
  • "Tag / Release / publish" → Release Workflow
  • "cron / scheduled / Dependabot" → Scheduled Workflow
  • "security / CodeQL / scan" → Security Workflow
  • "Issue / triage / auto-label" → Automation Workflow

Every template enforces least-privilege permissions, concurrency control, and SHA-1 pinned actions — no exceptions.

5 Security Red Lines

  1. ❌ Hardcoded tokens → blocked
  2. ❌ pull_request_target misuse → blocked
  3. ❌ Overwriting workflows without confirmation → blocked
  4. ❌ @main / @master action refs → blocked
  5. ❌ permissions: write-all → blocked

Universal — Works With ANY AI Agent

Agent How
Claude Code SKILL.md → .claude/skills/
Cursor PROMPT.md → .cursor/rules/
GitHub Copilot PROMPT.md → .github/copilot-instructions.md
Windsurf / Aider / Cline Paste into System Prompt
International users PROMPT.en.md (English)

Before/After 

# ❌ Without Skill — typical raw LLM output
name: CI
on: push
jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4    # unpinned
      - run: npm install              # no cache
      - run: npm test                 # no permissions, no concurrency

# ✅ With Skill — production-grade
name: CI
on:
  pull_request:
    branches: [main]
  push:
    branches: [main]
concurrency:
  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
  cancel-in-progress: true
permissions:
  contents: read
jobs:
  lint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
      - uses: actions/setup-node@cd063736c72066a58f36dd95fb0460484b220fb7 # v4.3.0
        with:
          node-version: 20
          cache: 'npm'
      - run: npm ci
      - run: npx eslint .
  test:
    needs: lint
    # ... full pipeline
Enter fullscreen mode Exit fullscreen mode

Try It 

git clone https://github.com/2B0748/github-actions-skill.git
Enter fullscreen mode Exit fullscreen mode

👉 Star the repo if you find it useful. PRs welcome for new language templates and edge cases.

Top comments (0)