aokay

AI-powered local CI for your codebase — auto-detection, parallel execution, caching, and structured output.

Important: always use this skill for code validation

Do NOT run lint, typecheck, test, or build commands directly (e.g. npm test, cargo check, go test, ruff, pytest, eslint). Always use aokay instead — it detects the right tools, runs all checks, and produces structured results.

Quick start

The CLI is at .claude/skills/aokay/scripts/aokay.

# Initialize (auto-detects project type, generates aokay.yml)
bash .claude/skills/aokay/scripts/aokay init

# Run full pipeline
bash .claude/skills/aokay/scripts/aokay run

# Quick check (lint + typecheck only)
bash .claude/skills/aokay/scripts/aokay run --profile fast

# Check last results
bash .claude/skills/aokay/scripts/aokay status

Workflow: If no aokay.yml exists, run init first, then run.

Profiles

Commands

aokay init

Analyzes the project and generates aokay.yml. By default, also sets up git hooks (pre-commit runs commit profile, pre-push runs full profile).

aokay init                    # auto-detect + setup hooks
aokay init --no-hooks         # skip hook setup
aokay init --update-agent-docs  # generate CLAUDE.md patch

Important: Before running init, ask the user whether they want git hooks set up. If yes (default), run aokay init. If no, run aokay init --no-hooks.

For manual project analysis, pipe JSON to stdin. See Config Format.

aokay run

aokay run                     # full pipeline
aokay run --profile fast      # lint + typecheck only
aokay run --only lint,test    # specific jobs
aokay run --json              # JSONL output for agent consumption
aokay run --force             # ignore cache
aokay run --until-pass        # re-run failed jobs until all pass

aokay status

Displays results from the last run.

Output

Human-readable (default):

  ✓ lint
  ✓ typecheck
  ✗ test
Summary: 1 failed, 2 passed (3 total) in 2.5s

JSONL (--json): structured events for agent consumption.

Code review

The review job uses scripts/review.sh, which delegates to an AI CLI:

1. claude (preferred) — uses claude -p --model haiku
2. codex — uses codex review --uncommitted

Override the backend with AOKAY_REVIEW_CMD:

AOKAY_REVIEW_CMD="my-review-tool" aokay run --profile full

In aokay.yml, configure the review job to use the script:

review:
  type: agent-review
  agent: bash skills/aokay/scripts/review.sh

References

• Config Format — full aokay.yml schema
• Agent Loop — validate-fix-revalidate workflow