About Workflows

Agentic workflows are AI-powered automation that understand context, make decisions, and take meaningful actions from natural language markdown instructions.

How It Works

Developers describe automation needs in plain language. GitHub Agentic Workflows runs markdown files as agents in GitHub Actions, executed by AI coding agents (Copilot CLI, Claude Code, Codex, etc.).

gh aw compile converts markdown into a hardened .lock.yml workflow embedding frontmatter and loading the markdown body at runtime. Compilation validates configuration, applies security hardening, and generates machine-ready YAML.

Use Cases

  • Issue triage, labeling, project coordination
  • CI failure diagnosis and test improvements
  • Documentation maintenance and consistency
  • Code simplification and refactoring
  • Daily reports and workflow health monitoring
  • Cross-repository feature sync

CI/CD Distinction

Agentic workflows complement — not replace — traditional deterministic CI/CD. Core build and release processes requiring strict reproducibility should remain conventional Actions workflows. Agentic workflows suit tasks benefiting from coding agent flexibility.

Early Development Warning

Documentation notes product is in early development, may change significantly, and even with supervision “things can still go wrong.”

Sample Workflow Structure

---
on:
  schedule: daily
permissions:
  contents: read
  issues: read
safe-outputs:
  create-issue:
    title-prefix: "[report] "
---
## Daily Report
Create a status report as a GitHub issue...

Triggers: new issues, issue comments, pull requests, PR comments, discussions, schedules, manual runs.