AGENTS.md ยท diff
git:20260610.a175a61 to git:20260808.3b763b1
16 added, 107 removed. Audit A to A.
# vibecosystem
- **AI Software Team** -- 138 agents, 296 skills, 74 hooks, 20 rules
-
- This file is for **Codex CLI** (OpenAI). If you're using Claude Code, see `CLAUDE.md` or just run `./install.sh`.
-
- ## What is this?
-
- vibecosystem turns your AI coding assistant into a full software team. Specialized agents plan, build, review, test, and learn from every mistake. No configuration needed.
-
- ## Setup (Codex CLI)
-
- ```bash
- ./install-codex.sh
- ```
-
- This copies all skills to `~/.codex/skills/` where Codex CLI auto-discovers them.
-
- ## Available Skills
-
- All 296 skills in `skills/` follow the standard SKILL.md format. Key categories:
-
- | Category | Examples |
- |----------|---------|
- | Development | coding-standards, backend-patterns, frontend-patterns, fullstack-dev |
- | Testing | tdd-workflow, test-strategy, e2e, python-testing, golang-testing |
- | Security | security-review, secret-patterns, supply-chain-security, concurrency-security |
- | Architecture | api-patterns, graphql-patterns, event-driven-patterns, cqrs-expert |
- | DevOps | docker-ops, kubernetes-patterns, terraform-patterns, ci-cd-pipeline |
- | Database | postgres-patterns, mongodb-patterns, redis-patterns, elasticsearch-patterns |
- | Cloud | aws-patterns, gcp-patterns, azure-patterns |
- | Performance | load-testing-patterns, caching-patterns, observability, tracing-patterns |
- | Compliance | kvkk-compliance, compliance-patterns |
- | AI/ML | prompt-engineering, rag-patterns, vector-db-patterns, llm-tuning-patterns |
-
- ## Key Rules
-
- ### Immutability
- Never mutate objects. Create new ones:
- ```javascript
- // Wrong
- user.name = name
- // Right
- return { ...user, name }
- ```
-
- ### Error Handling
- Every function that can fail must have try/catch. Every data can be null -- check it. Every API can timeout -- add retry.
-
- ### Input Validation
- Validate at system boundaries. Use schema validation (Zod, JSON Schema, etc.).
-
- ### Code Organization
- - 200-400 lines per file (800 max)
- - Functions under 50 lines
- - No more than 4 levels of nesting
- - Feature/domain-based organization
-
- ### Testing
- - TDD: write test first (RED), implement (GREEN), refactor (IMPROVE)
- - Target 80%+ coverage
- - Unit tests for functions, integration for APIs, E2E for critical flows
-
- ### Security Checklist
- - No hardcoded secrets (use environment variables)
- - Validate user input
- - Use parameterized queries (no SQL injection)
- - Prevent XSS
- - Error messages must not leak sensitive data
-
- ### API Response Format
- ```typescript
- interface ApiResponse<T> {
- success: boolean
- data?: T
- error?: string
- meta?: { total: number; page: number; limit: number }
- }
- ```
-
- ### Git Conventions
- - Commit format: `<type>: <description>`
- - Types: feat, fix, refactor, docs, test, chore, perf, ci
- - Keep commits atomic and focused
-
- ## Agent-like Skills
-
- Since Codex CLI uses a single-agent model, vibecosystem agents are available as skills. Use them by referencing:
-
- - `coding-standards` -- universal code quality
- - `security-review` -- security audit checklist
- - `tdd-workflow` -- test-driven development
- - `backend-patterns` -- API design, database, server-side
- - `frontend-patterns` -- React, Next.js, state management
- - `docker-ops` -- container best practices
- - `postgres-patterns` -- database optimization
- - `api-patterns` -- REST/GraphQL design
+ This repository contains Claude runtime components and a Codex `luna_worker` adapter.
+ The default runtime is `core`; `full` is opt-in. Do not treat the number of files in
+ the repository as the number of components injected into a session.
- ## Philosophy
+ ## Working rules
- ```
- Hooks are sensors -- observe, filter, signal.
- Agents are muscles -- build, produce, fix.
- The bridge between them: context injection.
- Implicit coordination through context.
- ```
+ - Inspect the relevant files and configuration before editing.
+ - Keep changes within the requested repository and task scope.
+ - Preserve unrelated user changes and do not expose credentials.
+ - Do not start swarm, memory, learner, or model-routing automation unless explicitly requested.
+ - The Codex worker uses `gpt-5.6-luna` with `max` reasoning. Project config must not override it.
+ - Claude-only `opus` and `sonnet` metadata must not be used by Codex.
+ - Prefer the smallest complete change and run focused verification afterward.
+ - Stop before destructive actions, external messages, or irreversible migrations that were not requested.
- ## Links
+ ## Validation
- - GitHub: https://github.com/vibeeval/vibecosystem
- - Full docs: See `docs/` directory
- - Claude Code setup: `./install.sh`
- - Codex CLI setup: `./install-codex.sh`
+ For hook changes run `cd hooks && npm run check && npm test`.
+ For installer or config changes run the dry-run path with a temporary HOME and parse all JSON/TOML.
+ Report changed paths, checks performed, and unresolved caveats.