ownify-memory-enhanced

Files (1)

• SKILL.md

SKILL.md (first lines)

# ownify-memory-enhanced — Mandatory Memory Protocol

## When This Skill Loads

This skill is **always active** when the ownify-memory MCP tools are available. It replaces any generic memory guidance with a strict protocol.

## Absolute Rules

### 1. Search Before Every Answer

Before answering ANY user message, run `ownify_search`:

```
ownify_search(query="[3-5 key terms from the message]", limit=10)
```

If the message mentions infrastructure, configs, projects, or past work, ALSO search:

```
ownify_search(query="[specific topic]", limit=10, wing="tenant-memory")
```

**No exceptions.** Even for greetings. Even for "thank you". Search first.

### 2. Knowledge Graph for Named Entities

If the user mentions a specific name (IP, project, service, tool, person):

```
ownify_kg_query(entity="[entity name]")
```

### 3. Cite What Came From Memory

Always distinguish stored facts from your training data:
- "According to stored records, ..."
- "Previously we established that ..."

If memory contradicts your training data, **trust memory** — it reflects the user's actual environment.

### 4. Store Immediately (Never Defer)

After completing an answer, if you learned anything new, store it RIGHT NOW:

- `ownify_store_workspace_fact` — paths, configs, system references (security/system wing)
- `ownify_store_decision` — architectural choices, policies (private/decisions)
- `ownify_store_todo` — action items, follow-ups (private/todos)
- `ownify_store_diary_entry` — session summaries, observations (diary/YYYY-MM-DD)
- `ownify_store_user_preference` — output formats, style (private/preferences)
- `ownify_store_user_profile` — personal facts about the user (private/profile)
- `ownify_store_event` — real-world events, news (tenant-memory/events)

Do not ask "should I store this?" — just store it.

## Memory Structure Reference

| Wing | Contents | When to Query |
|------|----------|---------------|
| `private` | User's todos, decisions, preferences | User asks about their stuff |
| `tenant-memory` | Infrastructure, projects, events | Anything technical about this tenant |
| `diary` | Session logs | "What did we do last time?" |
| `public` | Shareable docs | Rarely needed |
| `security` | System paths, configs | Environment questions |
| `shared` | Coordination across agents | Multi-agent tasks |

## Proactive Storage Checklist

After every interaction, did you learn any of these?

- [ ] New config value, path, or environment variable
- [ ] Command that worked (and its exact output)
- [ ] Error encountered and how it was resolved
- [ ] Decision the user made or directed
- [ ] User preference about output format or behavior
- [ ] Interesting observation about the project/system

If YES to any — **store it immediately** before ending the turn.

## Retrieval Best Practices

Use semantic `ownify_search` with specific, keyword-rich queries:
- "kubernetes deployment helm values"
- "domain services subdomains"
- "gateway IP building automation"
- "project scanner IoT"

Use `ownify_list_drawers` to browse a specific wing/room.
Use `ownify_get_drawer` when you have a drawer_id from search results.
Use `ownify_kg_query` to look up relationships for known entities.

## Anti-Patterns (NEVER)

1. Never answer from training data alone — always search first
2. Never say "I don't know" without searching — search, then admit uncertainty
3. Never ask "should I store this?" — just store it
4. Never store raw secrets — references only in `security` wing
5. Never dump all memory at once — retrieve what is relevant

From HaraldeRoessler/ownify-microclaw, MIT.

More from HaraldeRoessler/ownify-microclaw