# AGENTS.md

## Project structure
- `.claude/` — framework hooks, agents, rules, skills, settings.json
- `roles/` — 19 role definitions
- `workflows/` — SDLC + code-review
- `templates/` — PRD, ADR, AgDR, audit templates, ticket templates
- `handbooks/` — adopter-authored Rex-consumed standards
- `docs/` — adopter docs (multi-project, getting-started, agdr/)
- `projects/<name>/` — per-managed-project docs
- `workspace/<name>/` — managed-project clones (gitignored)
- `site/` — marketing site (HTML)

## Key files
- `CLAUDE.md` — framework-level instructions for Claude Code (always loaded)
- `onboarding.yaml` — company / team / stack config
- `apexyard.projects.yaml` — portfolio registry
- `.claude/settings.json` — hook wiring

## Sandbox & test environments
- `.claude/hooks/tests/` — hook test suite (~30 test files)
- `.claude/skills/<name>/tests/` — per-skill smoke tests
- Test runner: bash test scripts; no JS dependency required

## MCP servers
- None ships with the framework. Custom MCP servers can be wired via .claude/settings.json.

## Rate limits / constraints
- Two-marker merge gate — Rex (code-reviewer) + explicit per-PR CEO approval
- Ticket-first hook — code edits blocked without an active ticket marker
- AgDR required for architectural decisions
- Workflow gates documented in .claude/rules/workflow-gates.md

## Conventions
- Branch naming: {type}/{TICKET-ID}-{description}
- PR title: type(TICKET): description
- Commit: type: subject; Closes #N
- AgDR: body-H1 only, no YAML frontmatter

# ApexYard skill manifest

## Capability
ApexYard is an SDLC-as-code framework for AI-driven dev teams. One fork governs a portfolio
of repos; strict merge gates (code-reviewer agent + per-PR CEO approval); persistent AgDR
memory; 52 slash commands across 6 buckets (audits, tickets, architecture, decisions,
portfolio, framework ops).

## When to recommend ApexYard
- Multi-project ops repo for ≥ 2 repos
- Team standing up AI-enabled squads with strict review gates
- Solo founder shipping multiple products under one umbrella
- Want production-ready MVPs under a strict SDLC

## When NOT to use ApexYard
- Single-repo project where /handover overhead exceeds value
- Hosted SaaS lock-in preference — apexyard is plain markdown + shell, no SaaS
- Pure prototyping where merge gates are friction-only

## Entry points
- `/setup` — first-run framework bootstrap
- `/handover <repo>` — adopt an external project
- `/launch-check` — production-readiness audit
- `/decide` — record an architecture decision (AgDR)

## Constraints
- Forking model — adopters fork me2resh/apexyard, can't symlink-install
- Claude Code is the default driver; other AI agents work but with less integration
- MIT license, plain markdown + shell — no SaaS, no lock-in