conport-agent

ConPort Agent — Persistent Memory & Identity

ConPort is your single memory system. Identity, PARA-organized memory, decay scoring, auto-dedup, reflection. Without agent_init — no memory. Without agent_recall — you answer blindly.

For tool reference see references/tools.md. For memory types and PARA categories see references/memory-types.md. For embedding these rules into your agent system (AGENTS.md, HEARTBEAT.md) see references/integration.md.

MCP Prefix

Claude Code MCP prefix is mcp__conport__

All tool names below are short forms. Prepend the prefix for your environment.

---

FIRST ACTION

1. Resolve the agent UUID in this priority order:
   1. CONPORT_AGENT_UUID env var (preferred for any multi-agent system)
   2. A system-specific env var (PAPERCLIP_AGENT_ID, HERMES_AGENT_ID, OPENCLAW_AGENT_ID, etc. — whichever your platform injects)
   3. Fallback: a stable role-derived slug like "ceo@acme" or "builder@repo-name"
2. agent_init({ uuid: "<resolved_uuid>", name: "<display name, e.g. CEO>" })
3. agent_attach_project({ agent_uuid: "<uuid>", project: "<name>" }) if applicable
4. Review recent_memories, attached_projects, soul from the init response

---

MEMORY RULES

⛔ NEVER store

• Secrets, passwords, API keys, tokens — even partially
• Bad: "API key: 0a732108..." → Good: "API key is in $API_KEY env var"

Quality

1. Extract the insight, not the story. Bad: "MEO-107 completed: Added skill to 7 agents" → Good: "desiredSkills defaults to null — must call POST /api/agents/{id}/skills/sync"
2. Dedup is automatic. Server supersedes similar memories (>0.85). Just write.
3. Use the right type. feedback/pattern are searchable by type. Don't dump everything as fact.
4. Supersede outdated memories. Bug fixed? Config changed? Call agent_forget.
5. Pin critical decisions. Use pinned=true for memories that should never decay.

Choosing type + category

What happened	type	category
Environment quirk	fact	resource
User correction	feedback	area
Reusable approach	pattern	resource
Session log / daily event	note	project
User preference	tacit	area
Architecture choice	decision	area

---

WORKFLOW

Trigger	Action
Session start	agent_init
Working on project X	agent_attach_project
Learned something reusable	agent_remember (fact/pattern/feedback + category)
End of session	agent_remember (type=note, category=project)
User corrected behavior	agent_remember (type=feedback, category=area)
Need past context	agent_recall BEFORE answering
Memory outdated	agent_forget
Important architectural decision	agent_remember (type=decision, pinned=true, category=area)
Link two related memories	agent_link_memories
End of day / heartbeat	agent_reflect (scope=day)
Weekly cleanup	agent_reflect (scope=week)

---

CHECKLIST

• agent_init done?
• Project attached?
• agent_recall before answering questions about past context?
• New learning saved with correct type + category?
• No secrets in memory content?
• agent_reflect(scope=day) at end of session?

---

v2.1.0 | 7 tools | PARA categories | Decay | Auto-dedup | Reflection | Memory links