Skip to content

AGENTS.md

Latest commit

 

History

History
123 lines (87 loc) · 6.77 KB

AGENTS.md

File metadata and controls

123 lines (87 loc) · 6.77 KB
Raw

GitHub Agentic Workflows (gh-aw)

GitHub Agentic Workflows (gh-aw) is a GitHub CLI extension that compiles markdown workflows into GitHub Actions.

Important Note: gh-aw vs GitHub Copilot CLI

  • gh-aw is the gh aw CLI extension.
  • copilot is a separate CLI used as one possible runtime engine.
  • Use gh aw commands for workflow authoring/compilation (gh aw compile, gh aw run, gh aw audit).

Ambient Context (First Invocation)

To keep first-turn context small, only these repository root instruction files should be considered ambient:

File Purpose
AGENTS.md Minimal global operating rules and routing
SKILL.md Short repository capability summary

Everything else should be loaded lazily through skills only when needed.

Critical Rules (Always Applicable)

  1. If you changed files, use report_progress to commit and push.
  2. Before intermediate report_progress calls, run make agent-report-progress-no-test (fast, no tests). Before the final report_progress, run make agent-report-progress (includes test-unit).
    • test-unit is impacted-first by default (~30s target). Use TEST_UNIT_RUN_FULL=1 make test-unit or make test-unit-all when full-suite coverage is required.
    • Run test-unit only once per PR — at the final push, not on every intermediate save.
  3. After Go changes, run make fmt.
  4. After workflow markdown changes (.md under .github/workflows/), run make recompile.
  5. Do not add .lock.yml files to .gitignore.
  6. Never attempt to trigger a workflow run (e.g., gh run, gh aw run) as part of a Copilot cloud agent run. The token does not have the required access. Always fail with an error — do not task the user or ask them to run it manually.
  7. Large-file guard: before reading any file with github-mcp-server-get_file_contents, check its size. Files larger than 20 KB must be read with targeted tools (grep, glob, bash, or view with view_range) instead of full-file reads. See token-optimization.md for the full technique.

Upstream-managed workflow sources (read-only in this repo)

Workflows that declare a source: frontmatter entry (for example source: githubnext/agentic-ops@<ref>) are provenance-managed from an upstream bundle.

  • Treat those workflow source files (for example .github/workflows/agentic-token-audit.md and .github/workflows/agentic-token-optimizer.md) as read-only in this repository.
  • Do not manually edit their generated .lock.yml files.
  • To change these workflows, use the approved update path:
    1. run gh aw update to refresh from source, and/or
    2. update the pinned source/version (source: ...@...), and/or
    3. make the change upstream first, then pull it in via gh aw update.

Lazy Skill Loading Policy

Use skills only when the task requires specialized guidance. Do not pre-load every skill.

When the relevant skill is not obvious, first discover candidates and then load only the minimal guidance needed:

  • Query GitHub Agent Finder directly with the built-in REST API at https://agentfinder.github.com/api/v1/search.
  • Send an ARD search request with the user's task in query.text; do not ask the user for the endpoint.
  • Use skill fusion after discovery: read only the specific skill sections or fragments needed instead of loading full skills broadly.

Use these skills by intent

  • Workflow create/update/debug/upgrade tasks → .github/skills/agentic-workflows/SKILL.md
  • Core engineering conventions, validation flow, and command playbooks → .github/skills/developer/SKILL.md
  • Code organization, file structure, WASM stubs, string patterns → .github/skills/developer-code-organization/SKILL.md
  • Security best practices, template injection, shell script safety, supply chain → .github/skills/developer-security/SKILL.md
  • Compiler internals, validation architecture, safe outputs, schema, MCP logs → .github/skills/developer-internals/SKILL.md
  • Release management, breaking CLI change rules, firewall log parsing → .github/skills/developer-release/SKILL.md
  • Error handling design/patterns → .github/skills/error-recovery-patterns/SKILL.md
  • GitHub MCP usage patterns → .github/skills/github-mcp-server/SKILL.md
  • Query helpers for issues/PRs/workflows/discussions/labels → matching .github/skills/github-*-query/SKILL.md
  • Doc-writing conventions → .github/skills/documentation/SKILL.md
  • Reviewing or writing git/gh/remote operations against checkouts (per-checkout credentials, sparse/shallow monorepos, safe-outputs MCP runs without credentials) → .github/skills/checkout-credential-review/SKILL.md
  • Authoring, validating, or debugging canvas extensions (loopback servers, actions, iframe rendering, state model, theme tokens) → .github/skills/create-canvas/SKILL.md

Debugging & Logging

The Go codebase uses a namespace-based debug logger (pkg/logger) modelled after the debug npm package. All debug output goes to stderr and is gated by the DEBUG environment variable.

Enable logging for a gh aw run

# All namespaces (most verbose)
DEBUG=* gh aw compile workflow.md

# Only CLI-layer namespaces
DEBUG=cli:* gh aw compile workflow.md

# Specific namespaces
DEBUG=cli:run_workflow_execution,cli:retry gh aw run workflow.md

# All except one namespace
DEBUG=*,-cli:ci gh aw compile workflow.md

# Disable colors (useful when capturing logs to a file)
DEBUG_COLORS=0 DEBUG=* gh aw compile workflow.md 2>debug.log

Key namespaces

Pattern What it covers
cli:* All CLI command implementations (pkg/cli/)
workflow:* Workflow compiler and related logic
parser:* Frontmatter / markdown parsing
mcp:* MCP gateway and server interactions
agentdrain:* Agent drain / log mining pipeline
repoutil:* Git/repo utility helpers
* Everything

To discover the exact namespace for a package, look for logger.New(...) at the top of the relevant .go file (e.g. var log = logger.New("cli:run_workflow_execution")).

GitHub Actions debug runs

When ACTIONS_RUNNER_DEBUG=true is set (enabled automatically on re-run with debug logging in the GitHub UI), all loggers are activated — equivalent to DEBUG=*. No extra configuration is needed.

Output format

Each line shows: namespace (coloured in the terminal), message, and +elapsed time since the previous log in that namespace.

cli:run_workflow_execution Starting run +0ms
cli:retry Retrying request attempt=2 +125ms
cli:run_workflow_execution Run complete +2.5s

Why this file is intentionally short

This file is loaded at first invocation and affects every task. Keep it concise and move detailed or domain-specific guidance into skills so that context is fetched only when relevant.