| name | memo-brain |
| description | Manage and retrieve cross-conversation memory. Public action semantics follow command-philosophy. Use for "remember this", "search memory", "organize memory", or "show current state". |
Memo Brain Management
This skill follows the public action language defined in docs/architecture/command-philosophy.md.
Standard Actions
memo awaken
memo remember
memo recall
memo reflect
memo dream
memo state
Current Capability Boundaries
The current CLI does not provide these old interfaces. Do not reason or act as if they still exist:
memo embed
memo search
memo restore
memo update
memo merge
memo delete
memo list
--tags
--after / --before
If the user speaks in that old product language, translate it into the standard action semantics. If the current system cannot support the request, say so directly instead of fabricating capabilities. search / embed may appear only as natural-language triggers or warnings about old terminology, not as runnable command examples.
When to Use
Use this skill when:
- The user explicitly asks to remember or record something
- The user wants to search or recall past memory
- You need details for one memory record
- You need to run dream / maintenance
- You need current engine state
- You need dream-driven derived-layer maintenance
Do not use this skill when:
- The task is ordinary code search inside the repository
- The task does not need cross-conversation memory
- The request depends on update/merge/delete/list behaviors that do not exist in the current CLI
Recommended Workflow
Awaken a Memory Space
memo awaken
Remember Content
Standard action: remember
memo remember "<content>"
If you already know structured information, add it explicitly:
memo remember "<content>" --entity person:Alice --entity place:Paris --fact Alice:lives_in:Paris
Recall Content
Standard action: recall
memo recall "<query>" -n 10
If fast-path retrieval is likely insufficient:
memo recall "<query>" -n 10 --deep
Reflect on One Memory
Standard action: reflect
memo reflect <memory-id>
Dream
Standard action: dream
memo dream
For full derived-layer maintenance:
memo dream --full
Inspect State
Standard action: state
memo state
How to Choose the Action
| User Intent | Standard Action | Current Execution |
|---|
| "remember this conclusion" | remember | memo remember ... |
| "did we solve something like this before" | recall | memo recall ... |
| "show me that memory in detail" | reflect | memo reflect ... |
| "organize the memory" | dream | memo dream |
| "what is the current system state" | state | memo state |
| "indexes may be inconsistent, restore them" | dream | memo dream --full |
Retrieval and Recording Principles
Recording Principles
- Record durable experience, facts, decisions, or troubleshooting outcomes worth keeping
- Focus on the content first; add
--entity and --fact when you can do so concretely
- Do not design workflows around nonexistent features such as tags, update, merge, or list
- Default
remember does not call providers; do not make embedding writes the default recording premise
Retrieval Principles
- Queries should include situation and intent, not only loose keywords
- Start with default
memo recall
- Only use
--deep when default results look weak, the topic spans multiple layers, or the user explicitly asks for deeper recall
- Use
memo reflect when you need detail on one returned record
- Default
recall does not call providers; do not claim full semantic retrieval when providers are absent
Maintenance Principles
- Derived-layer maintenance goes through
memo dream; use memo dream --full only when full maintenance is needed
dream may enter extraction / embedding slow paths; do not present those slow paths as default remember / recall behavior
- Use Working Set / Pinned for user-facing memory semantics; do not expose L0/session as the user's mental model
Common Mistakes
| Don't | Do |
|---|
Keep calling memo search / memo embed | Translate into standard action semantics |
| Pretend old commands are still the standard | Use awaken/remember/recall/reflect/dream/state directly |
| Fake update/merge/delete/list capabilities | State directly that they are not implemented in the current CLI |
Treat extract as the main memory entrypoint | Organize the workflow around public actions only |
Treat memo restore as standard maintenance | Use memo dream, or memo dream --full when needed |
| Explain user-visible state with L0/session | Use Working Set / Pinned |
Trigger Phrases
| Action | Trigger Phrases |
|---|
remember | "remember this", "record this", "save this experience" |
recall | "how did we do it before", "search memory", "do you remember" |
reflect | "show that memory in detail", "open the details" |
dream | "organize memory", "run dream", "restore derived layers", "restore index state" |
state | "show current state" |
For executable examples, see examples.md.