Entry point for agents. Binding behavior:
AGENTS.md. Skill router:SKILLS.md.
Recommended behavior:
Every backend task
-> AGENTS.md
-> SKILLS.md (pick skills)
-> backend-core playbook + skills/backend/CORE.md
-> only the relevant skill(s): playbook + reference
-> backend-testing before final validation
Structure:
AGENTS.md # Global agent instructions (§§1–14)
SKILLS.md # Skill router: which skills, loading order, quick picks
CLAUDE.md / GEMINI.md # Adapters (pointers, not source of truth)
.github/copilot-instructions.md # Copilot adapter (pointer)
.agents/skills/<skill>/SKILL.md # Playbooks (HOW): steps, code shapes, checklists, common mistakes
skills/backend/*.md # References (WHAT/WHY): verified facts + binding rules (CORE.md, API.md, ...)
docs/agent/ # End-to-end CRUD recipe, known issues, project map
tasks/TASK_TEMPLATE.md # Canonical task template
tasks/README.md # Task workflows (instances in docs/tasks/)
docs/tasks/ # Task instances
Examples:
New CRUD resource
-> backend-api + backend-data + backend-security + backend-testing
+ docs/agent/STANDARD_CRUD_SERVICE_REPOSITORY.md
Fix refresh token security
-> backend-security + backend-testing
Add RabbitMQ consumer
-> backend-async-messaging + backend-data (if persistence involved) + backend-testing
Fix file streaming
-> backend-files + backend-security + backend-testing
Ordinary user CRUD
-> DO NOT read backend-ai-rag
RAG ingestion change
-> backend-ai-rag + backend-data + backend-files + backend-testing as applicable
See SKILLS.md for the full routing table and tasks/README.md for task creation/execution.