An interactive book on Continuous AI
GitHub Agentic Workflows
Write your repository's outer loop in plain Markdown. Compile it to a hardened, SHA-pinned GitHub Actions workflow, and let a coding agent do the triage, docs, and review work — safely, and reviewed by default.
How to read this book
One running example, the Repo Assistant, grows from a single triage workflow into a governed multi-repo fleet. The chapters build in three parts that widen in scope. Two kinds of reader travel the same pages, guided by margin notes.
Builders author, compile, and run. Read in order, work every Worked example, and follow the Builder notes.
Leaders weigh safety, cost, and adoption. Skim the concept and when to use sections and follow the Leader notes; Parts Two and Three are written for you.
Contents
14 chapters across 3 parts
The Individual
one workflow- 01 What Are Agentic Workflows? Explain what an agentic workflow is, why the outer loop matters, and when to reach for gh-aw instead of plain GitHub Actions. 10 min
- 02 The 10-Minute Win: Your First Workflow Install the gh aw CLI and ship a first working Repo Assistant that triages a new issue end to end. 15 min
- 03 Anatomy & the Compile Model Read any workflow's frontmatter + Markdown, run the compile-and-iterate loop, and understand what the generated .lock.yml contains. 9 min
- 04 Triggers: When Workflows Wake Up Choose the right on: events so the Repo Assistant runs at exactly the right moments and no others. 9 min
- 05 Engines: Choosing the Agent's Brain Select and configure an engine (Copilot, Claude, Codex, or Gemini) and understand the portability that engine-neutral design buys you. 7 min
The Team
safe, reviewed, patterned- 06 Safe Outputs: Acting Without Overreach Let the Repo Assistant write to the repo — issues, comments, PRs — through the sanitized safe-outputs boundary instead of raw permissions. 7 min
- 07 Defense in Depth: Permissions, Firewall & Strict Mode Harden a workflow with least-privilege permissions, an egress firewall, and strict mode so a compromised prompt can do little damage. 7 min
- 08 Tools & MCP: Real Capabilities, Governed Give the Repo Assistant real capabilities with the tools: block and MCP servers while keeping every capability governed. 7 min
- 09 Continuous Triage & Docs: Reading the Room Ship two production-shaped patterns — Continuous Triage and Continuous Docs — as mini-products the Repo Assistant runs on its own. 6 min
- 10 Continuous Review, Testing & CI-Doctor Close the quality loop with Review, Testing, CI-Doctor, and Refactoring patterns while keeping humans on the merge decision. 6 min
The Organization
fleet at scale- 11 Reuse & Memory: Shared Components and Repo Knowledge Factor common intent into imported shared components and give the Repo Assistant memory that persists across runs. 8 min
- 12 Trust & Operate: Observability and Debugging Inspect, debug, and audit runs with gh aw logs, gh aw audit, and OpenTelemetry so you can trust what the fleet does. 7 min
- 13 Governance & FinOps: Policy and Cost at Scale Cap, meter, and gate agentic spend with AI Credits and max-ai-credits, and set org policy so the fleet stays affordable and compliant. 8 min
- 14 Fleets & Adoption: From One Repo to the Org Scale the Repo Assistant into a governed multi-repo fleet and follow an enterprise adoption playbook to roll it out. 6 min