AI agent persona management, safety verification, memory sync, and identity rollback for Claude Code.
Powered by Soul Spec — the open standard for agent identity.
# Two commands. That's it.
/plugin marketplace add https://github.com/clawsouls/clawsouls-claude-code-plugin
/plugin install clawsouls
/reload-plugins| Command | Description |
|---|---|
/clawsouls:browse |
Search and install personas from ClawSouls |
/clawsouls:install |
Apply a persona to your project |
/clawsouls:scan |
Verify persona safety (53 patterns, A+ to F) |
/clawsouls:rollback |
Detect drift & restore identity |
/clawsouls:memory |
Git-based memory sync & TF-IDF search |
# Browse available personas
/clawsouls:browse coding agents
# Install a persona
/clawsouls:install clawsouls/surgical-coder
# Verify safety
/clawsouls:scan
# Search memory (TF-IDF ranked)
/clawsouls:memory search "SDK version fix"
# Sync memory to remote Git
/clawsouls:memory syncThe plugin connects a clawsouls-mcp server with 9 tools:
| Tool | Description |
|---|---|
soul_search |
Search personas by keyword, category, tag |
soul_get |
Get detailed persona info |
soul_install |
Download persona → generate CLAUDE.md |
| Tool | Description |
|---|---|
soul_scan |
SoulScan — 53 safety patterns, grade A+ to F |
soul_rollback_check |
Compare current vs. baseline, detect drift |
| Tool | Description |
|---|---|
memory_search |
TF-IDF + BM25 ranked search across markdown files |
memory_detail |
Fetch full content of a specific section |
memory_status |
File inventory + sizes + git status |
memory_sync |
Multi-agent Git sync (init/push/pull/status) |
No embedding model needed. No API calls. Pure local keyword intelligence.
memory_search query="authentication bug" → compact index with relevance scores
memory_detail file="memory/2026-03-31.md" line=15 → full section content
3-Layer Workflow for token efficiency (~10x savings):
memory_search→ compact index (~50 tokens/result)memory_detail→ full section for interesting resultsmemory_search enhanced=true→ deep dive with full snippets
Share memory across multiple agents via Git:
# Initialize remote
memory_sync action=init repo_url=git@github.com:user/agent-memory.git
# Push changes
memory_sync action=push agent_name=brad
# Pull from other agents
memory_sync action=pullCompatible with OpenClaw memory layout:
MEMORY.md # Long-term curated memory
memory/
topic-*.md # Project-specific context
YYYY-MM-DD.md # Daily logs
- Persona Manager — Activates when discussing agent identity or Soul Spec files
- SoulScan — Activates when checking safety or after persona modifications
- Swarm Memory — Activates when managing memory files or syncing knowledge
- Soul Rollback — Activates when behavior seems inconsistent with persona
- Persona Advisor — Expert help designing effective agent personas (
/agents→ persona-advisor)
claude --print --plugin-dir clawsouls-claude-plugin \
"Review the changes in this PR for security issues"claude --output-format stream-json --plugin-dir clawsouls-claude-pluginclaude --resume <session-id>
# Soul Rollback activates automatically if drift is detected| Platform | How |
|---|---|
| Claude Code | This plugin — /clawsouls:install |
| OpenClaw | Native SOUL.md support |
| Cursor / Windsurf | MCP server via .mcp.json |
| Any MCP Client | npx -y clawsouls-mcp@latest |
MIT