AI Documentation

AI-facing collaboration guidance lives here unless a tool requires a specific path.

Contents

• autonomous-bootstrap.md — single startup entrypoint for Hybrid AWP with Explicit Fallback Gate + local-only spec-injector workflow.
• claude-codex-cheatsheet.md — quick reference for Claude Code and Codex collaboration.
• claude-codex-workflow.md — full workflow guide for low-token Claude Code usage.
• codex-autonomous-workflow.md — autonomous worker profiles, routing rules, review gates, and PR scope contract.
• autonomous-pr-gates.md — autonomous evidence gates, review triage refs, and spec workflow-check boundaries.
• autonomous-threshold-policy.md — #664 calibrated controller-direct / worker-routing threshold policy.
• code-review-refactor.md — local Claude Code review workflow notes.
• github-ssh-443-push.md — playbook for GitHub SSH over 443 when git push is unstable.
• github-actions-debugging.md — playbook for PR, CI, CI scope router, and auto-ready debugging.
• openspec-workflow.md — OpenSpec SDD workflow for feature proposal, specs, design, tasks, and archive/sync expectations.
• supply-chain-security.md — dependency install, AI-agent package use, and developer-machine persistence guardrails.
• token-budget.md — token budget guidance for AI-assisted work.
• workflow-metrics-baseline.md — baseline GitHub metadata metrics for AI-native repo workflow friction and gate-weight audits.
• workflow-gate-weight-audit.md — #881-based audit of PR governance gate weight and Track A exit criteria.
• workflow-scope-threshold-sample.md — #898 sample of recent scope-violation / scope-exception PRs and threshold calibration recommendations.

Root-Level Exceptions

These files and directories stay outside docs/ai/ because tools discover them by convention:

• CLAUDE.md — Claude Code entrypoint and repo-specific instructions.
• AGENTS.md — Codex/agent entrypoint and repo-specific instructions.
• .claude/ — Claude Code commands, rules, and shared settings.
• .codex/ — Codex local configuration.
• .cursor/ — Cursor rules.

Root-level entrypoints should link to this directory when they need longer-form AI guidance.