菜单
Use when creating, editing, restructuring, or reviewing AGENTS.md files or the CLAUDE.md symlink. Covers Pawrrtal's two-tier layout, what belongs in AGENTS.md vs memory vs skills, and line budgets.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when changing, extending, testing, or reviewing the Bun-backed Paw CLI package and its generated agent skills.
Use when writing or reviewing Pawrrtal Effect TypeScript code in backend-ts: API contracts, HttpApi handlers, services, repos, layers, tagged errors, Effect tests, streams, SQL, auth middleware, or vendor-source API checks.
Use when running the Bun-backed Paw CLI: local health, active context, shell completions, output modes, profiles, and verification of the supported command surface.
Turn a bad or long Pawrrtal work session into durable project learning. Use when the user says $retro, asks to learn from a session, says too much went wrong, or asks to create/update skills or rules from recent failures.
Use when touching Pawrrtal channels, providers, tools, plugins, subagents, context providers, turn orchestration, or code that decides where an integration should live. Enforces the split between generic kernel code, manifest plugins, trusted runtime adapters, provider adapters, channel adapters, and agent runtime primitives.
Operate Pawrrtal as a live service: prove the running commit, local origins, Cloudflared public URL, database schema, Telegram channel, and shutdown state. Use when checking whether Pawrrtal is live, deployed, reachable, healthy, broken in Telegram, safe to restart, or safe to take down.
| name | agents-md |
| description | Use when creating, editing, restructuring, or reviewing AGENTS.md files or the CLAUDE.md symlink. Covers Pawrrtal's two-tier layout, what belongs in AGENTS.md vs memory vs skills, and line budgets. |
| File | Role |
|---|---|
Root AGENTS.md | Stub (~10 lines) → points to .agent/AGENTS.md |
.agent/AGENTS.md | Entry contract — bootstrap, non-negotiables, memory map |
CLAUDE.md | Symlink to root stub |
Nested AGENTS.md | Package-local overrides |
| Surface | Holds |
|---|---|
.agent/AGENTS.md | Read order, hard safety lines, where depth lives, brain map |
memory/personal/PREFERENCES.md | How the user works — taste, workflow, multi-agent |
memory/semantic/DOMAIN_KNOWLEDGE.md | Stable facts — ports, commands, structure, auth |
memory/semantic/LESSONS.md | Graduated lessons (learn.py / graduate.py) |
memory/semantic/DECISIONS.md | Major brain/architecture choices |
skills/* | Procedures (repo-operations, code-quality, …) |
rules/** | Path-scoped file traps (paths: globs) |
.cursor/plugins/pawrrtal/rules/ | Cursor .mdc (clean-code, github-actions) |
protocols/permissions.md | Tool safety (non-negotiable) |
Do not put preferences or commands in AGENTS.md — that duplicates memory. AGENTS.md only points to them.
.agent/AGENTS.md: ≤ 120 lines (orchestration only)DOMAIN_KNOWLEDGE.md: facts, no tastePREFERENCES.md: taste and workflow, no ports.agent/AGENTS.mdNot in AGENTS.md: full command lists, Always/Ask/Never essays, code examples, user UI taste.
DOMAIN_KNOWLEDGE.mdPREFERENCES.mdlearn.py or graduate.pyrepo-operations, etc.)wc -l AGENTS.md .agent/AGENTS.md memory/personal/PREFERENCES.md memory/semantic/DOMAIN_KNOWLEDGE.md
| Don't | Do |
|---|---|
| Paste preferences into AGENTS.md | PREFERENCES.md |
| Paste commands/ports into AGENTS.md | DOMAIN_KNOWLEDGE.md |
Duplicate in workspace-context skill | Point at memory files |
| Grow root stub | Edit .agent/ + memory |