一键导入
agent-protocol
Specification for making MUSE role files machine-readable as an Agent Protocol — enabling multi-agent coordination and tool interoperability
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。
菜单
Specification for making MUSE role files machine-readable as an Agent Protocol — enabling multi-agent coordination and tool interoperability
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 about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always. Upgraded with AC-first workflow, pre-flight fast-fail, and structured Judge verdicts (inspired by opslane/verify).
Guide for creating Claude Code skills following Anthropic's official best practices. Use when user wants to create a new skill, build a skill, write SKILL.md, or needs skill creation guidelines. Provides structure, naming conventions, description writing, and quality checklist.
Pattern for progressively refining context retrieval to solve the subagent context problem
Fast iteration with two-stage review (spec compliance, then code quality)
Multi-agent coordination SOP derived from Anthropic's internal coordinator architecture. Use when orchestrating parallel workers, dispatching subagents, planning multi-step implementations, or when tasks require research→synthesis→implementation→verification workflow. Triggers: 'coordinate', 'dispatch workers', 'parallel agents', 'multi-agent', 'fan out', 'orchestrate'.
Use to run multiple subagents concurrently on independent tasks
| name | agent-protocol |
| description | Specification for making MUSE role files machine-readable as an Agent Protocol — enabling multi-agent coordination and tool interoperability |
Inspired by MemOS (multi-agent isolation + sharing) and multi-agent-memory (cross-agent persistence). Adapted for MUSE's pure Markdown zero-dependency architecture. Added in v2.12.0 (Batch 2 — competitive tech absorption).
MUSE role files (.muse/*.md) currently work because AI Agents read Markdown and follow instructions. But as the ecosystem evolves toward multi-agent workflows (MCP servers, Claude Projects, multi-tool chains), role files need a machine-readable protocol — not just human-readable documents.
Problems this solves:
.muse/
├── strategy.md # Strategy role
├── build.md # Build/Dev role
├── growth.md # Growth/Marketing role
├── qa.md # QA/Verification role
├── gm.md # General Manager role
├── ops.md # Operations role
├── research.md # Research role
├── fundraise.md # Fundraise role
└── archive/ # Archived content (not loaded)
Discovery rule: Any .md file in .muse/ (excluding archive/) = a role file.
Every role file MUST have an L0 comment as its first line:
<!-- L0: {version} | {top_priority} | {status_flags} -->
Parsed fields:
| Field | Type | Description | Example |
|---|---|---|---|
version | semver | Current project version | v2.11.0 |
top_priority | string | Most important current task | P0=竞品技术吸收Batch2 |
status_flags | csv | Comma-separated status indicators | QA PASS, 3 pending directives |
L0 is the Agent's "at a glance" API — any tool can grep all L0 lines to get project status in ~400 tokens.
Role files use Markdown headers as semantic sections. Tools should parse these:
| Section Header Pattern | Semantic Meaning | Agent Action |
|---|---|---|
## 📐 职责边界 | Role boundaries table | Respect: only act within listed responsibilities |
## ⏳ 待办 / ## 🎯 TODO | Pending tasks | Read for context; update checkboxes when completing |
## 📡 已接收战略指令 | Received directives | Check for pending work from other roles |
## 📡 QA→BUILD 通知 | Cross-role notifications | Process before starting new work |
## 🛠️ 技术栈 | Technology constraints | Respect in all code generation |
## 📋 CHANGELOG 维护规则 | Process rules | Follow during releases |
Directives are the message-passing API between roles:
📡 **S{NNN}→{TARGET}**: {STATUS} **{TITLE}**
> {BODY}
| Field | Format | Example |
|---|---|---|
S{NNN} | Sequential ID | S041 |
{TARGET} | ROLE or PROJECT/ROLE | MUSE/BUILD, GROWTH |
{STATUS} | 🟡 (pending) / ✅ (received) | 🟡 = needs pickup |
{TITLE} | One-line summary | 竞品深度调研回传 |
{BODY} | Blockquote detail | Full directive content |
Pickup protocol:
/resume scans for 🟡 directives📡 已接收战略指令 sectionmemory/
├── YYYY-MM-DD.md # Daily log (short-term, raw)
├── archive/ # Old logs
MEMORIES.md # Long-term knowledge (distilled + auto-captured)
| Memory Type | File | Write Method | Read Method |
|---|---|---|---|
| Short-term | memory/YYYY-MM-DD.md | /bye Step 4 | /resume Step ② |
| Long-term | MEMORIES.md | /distill (batch) + /bye Step 4.7 (auto-capture) | /resume Step ① |
| Constitutional | CLAUDE.md | Manual | Always loaded |
| User prefs | USER.md | /settings | /resume Step ④ |
┌─────────────────────────────────────────────┐
│ Agent Session (e.g., /resume build) │
│ │
│ ✅ CAN READ: │
│ - Own role file (.muse/build.md) │
│ - L0 lines of ALL role files │
│ - memory/*.md │
│ - MEMORIES.md, CLAUDE.md, USER.md │
│ - Source code & docs (on demand) │
│ │
│ ✅ CAN WRITE: │
│ - Own role file │
│ - memory/YYYY-MM-DD.md │
│ - MEMORIES.md (auto-capture only) │
│ - Source code (if build/dev role) │
│ │
│ ❌ MUST NOT: │
│ - Write to other role files directly │
│ - Create files outside .muse/ convention │
│ - Discuss other roles' topics in depth │
│ │
│ 📡 CROSS-ROLE COMMS: │
│ - Only via directive protocol (📡 S{NNN}) │
│ - Sender writes directive to OWN file │
│ - Receiver picks up during /resume │
└─────────────────────────────────────────────┘
For users managing multiple projects:
~/ProjectA/.muse/build.md ← ProjectA's build role
~/ProjectB/.muse/build.md ← ProjectB's build role
~/ProjectA/.muse/strategy.md ← May hold cross-project directives
Cross-project directives use PROJECT/ROLE targeting:
📡 S041→MUSE/BUILD: 🟡 ... ← Targets MUSE project's BUILD role
📡 S040→DYA/GROWTH: ✅ ... ← Targets DYA project's GROWTH role
Path resolution: CLAUDE.md MAY specify absolute paths for cross-project strategy files:
> strategy.md 绝对路径: `/Users/jj/Desktop/DYA/.muse/strategy.md`
An MCP server implementing MUSE protocol is included — see scripts/mcp-server.sh (added in v2.15.0).
| Tool | Description | Status |
|---|---|---|
muse_get_status | Read all L0 lines → return project status | ✅ |
muse_list_roles | List role files with summaries | ✅ |
muse_get_role | Read a specific role file (L1) | ✅ |
muse_send_directive | Create a 📡 directive in a role file | ✅ |
muse_write_memory | Append to today's memory file | ✅ |
muse_search_memory | Search across memory files + MEMORIES.md | ✅ |
Configuration: Add to your MCP client config:
{
"mcpServers": {
"muse": {
"command": "/path/to/muse/scripts/mcp-server.sh",
"args": ["--project-root", "/path/to/project"]
}
}
}
# Get project status (all L0 lines)
grep "<!-- L0:" .muse/*.md
# Check for pending directives
grep "🟡" /path/to/strategy.md
# Count role file lines (bloat check)
wc -l .muse/*.md
| Version | Date | Changes |
|---|---|---|
| v1.1 | 2026-03-15 | MCP server implemented (scripts/mcp-server.sh) |
| v1.0 | 2026-03-15 | Initial spec (Batch 2) |