openmemory

OpenMemory — OpenMemory MCP Operation Playbook

1. Overview

This skill is the operational playbook for OpenMemory (MCP) — a primary persistent memory system; the file-based fallback is a secondary index only (one-line pointers, no full-text copies). Configure project_id based on your workspace (see §4 Step 2).

---

2. HARD-GATE

---

3. Flow Router

---

4. SAVE Flow (5 Steps)

Step 1 — Classify sector

Step 2 — Choose scope

Default bias: project scope (openmemory_store_project).

Note: Set {{PROJECT_ID}} to your actual project identifier (e.g. "my-project", "airflow-agent", "data-platform"). This is a per-workspace configuration.

Step 3 — Choose type

Step 4 — Compose tags

Attach 2–5 lowercase hyphen-separated tags. Examples:

• deployment, rollback, incident
• api, dependency, architecture
• user-preference, feedback
• pattern, decision, architecture
• workflow, ci-cd, config

Step 5 — Execute

1. Call the appropriate MCP tool (openmemory_store_project or openmemory_store) with sector, type, tags, and content.
2. Optionally add one line in a local index file pointing to the stored memory (never copy the full content).

---

5. QUERY Flow (4 Steps)

Step 1 — Determine query strategy

Step 2 — Set parameters

Step 3 — Execute query

Call openmemory_query with the determined parameters.

Step 4 — Reinforce if used

If the retrieved memory is directly used in the response, call openmemory_reinforce with a boost of 0.05–0.1 to strengthen its salience for future sessions.

---

6. MAINTAIN Flow (3 Steps)

Step 1 — Scan

Call openmemory_list with:

• project_id: "{{PROJECT_ID}}" (omit for global scope)
• sector filter (optional)
• limit: 20 (adjust for larger scans)

Step 2 — Evaluate each entry

Step 3 — Report

Summarize to the user: what was deleted, reinforced, or corrected, and why.

---

7. Proactive Behaviors

Pre-task query

Before a non-trivial task, silently query OpenMemory for relevant context. Do not announce the query itself — only inform the user if the retrieved memory would affect the task's direction or decisions.

Post-task save

When the task produces a stable conclusion (not intermediate noise), save it. Briefly inform the user:

"已存入 openmemory: [one-line summary]"

Correction on discovery

If you notice an existing stored memory is factually wrong, immediately:

1. Delete the incorrect entry.
2. Save the corrected version.
3. Inform the user: "已更新 openmemory: [what changed]"

---

8. Decision Flow

digraph memory_flow {
    rankdir=LR;
    node [shape=box, style=rounded, fontname="sans-serif"];
    edge [fontname="sans-serif", fontsize=10];

    trigger [label="🔄 Trigger\n(user signal /\nagent discovery)", shape=ellipse, style=filled, fillcolor="#E8F0FE"];
    route  [label="Route?", shape=diamond, style=filled, fillcolor="#FFF3CD"];

    save   [label="SAVE\n1. Classify sector\n2. Choose scope\n3. Choose type\n4. Compose tags\n5. Execute", fillcolor="#D4EDDA"];
    query  [label="QUERY\n1. Strategy\n2. Parameters\n3. Execute\n4. Reinforce", fillcolor="#CCE5FF"];
    maint  [label="MAINTAIN\n1. Scan\n2. Evaluate each\n3. Report changes", fillcolor="#F8D7DA"];

    done   [label="Done", shape=doublecircle, style=filled, fillcolor="#E2E3E5"];

    trigger -> route;
    route  -> save   [label="SAVE signal"];
    route  -> query  [label="QUERY signal"];
    route  -> maint  [label="MAINTAIN signal"];

    save   -> done;
    query  -> done;
    maint  -> done;
}

---

9. Constraints (Anti-Pattern Table)