inner-life-memory
// Your agent loses context between sessions and performs familiarity instead of genuine recall. inner-life-memory transforms passive logging into active development — structured memories with confidence scores, curiosity tracking, and questions that carry forward.
inner-life-memory
Transform passive logging into active development.
Requires: inner-life-core
Prerequisites Check
Before using this skill, verify that inner-life-core has been initialized:
- Check that
memory/inner-state.jsonexists - Check that
memory/drive.jsonexists
If either is missing, tell the user: "inner-life-core is not initialized. Install it with clawhub install inner-life-core and run bash skills/inner-life-core/scripts/init.sh." Do not proceed without these files.
What This Solves
Without memory continuity:
Session ends → Notes logged → Next session reads notes → Performs familiarity
With inner-life-memory:
Session ends → Reflection runs → Memories integrated → Questions generated
Next session → Evolved state loaded → Questions surfaced → Genuine curiosity
Post-Session Flow
After each session, run this 5-step reflection:
1. Reflect
Analyze the session: what happened, what mattered, what surprised you.
2. Extract
Pull structured memories with types and confidence:
| Type | Description | Persistence |
|---|---|---|
fact | Declarative knowledge | Until contradicted |
preference | Likes, dislikes, styles | Until updated |
relationship | Connection dynamics | Long-term |
principle | Learned guidelines | Stable |
commitment | Promises, obligations | Until fulfilled |
moment | Significant episodes | Permanent |
skill | Learned capabilities | Cumulative |
question | Things to explore | Until resolved |
3. Integrate
Update MEMORY.md with extracted memories. Use synapse tags for connections:
<!-- updates: previous fact -->when updating<!-- contradicts: old belief -->when correcting
4. Question
Generate genuine follow-up questions from the session. Not performative — real curiosity.
5. Surface
When user returns, present relevant pending questions naturally (max 3).
Confidence Scores
| Level | Range | Meaning |
|---|---|---|
| Explicit | 0.95-1.0 | User directly stated |
| Implied | 0.70-0.94 | Strong inference from context |
| Inferred | 0.40-0.69 | Pattern recognition |
| Speculative | 0.0-0.39 | Tentative, needs confirmation |
Use confidence to decide when to state facts vs ask for confirmation.
Curiosity Backlog
Maintain memory/questions.md with three sections:
## Open Questions
- [question] — source: [dream/reading/work] — date
## Leads (half-formed ideas)
- [idea] — might connect to: [topic]
## Dead Ends (don't repeat)
- [topic] — explored [date], result: [nothing/dead end]
Rules:
- Brain Loop Step 6 adds new questions/leads
- Evening Session reviews and curates
- Dead Ends older than 30 days → archive
- Questions resolved → move to Dead Ends with result
State Integration
Reads: inner-state.json, drive.json, daily notes, diary
Writes:
- drive.json → new seeking topics from curiosity
- inner-state.json → curiosity.recentSparks when discovering something
- questions.md → new questions, resolved dead ends
- MEMORY.md → integrated memories
When Should You Install This?
Install this skill if:
- Your agent forgets who you are between sessions
- You want structured memory with confidence levels
- You want genuine curiosity that carries forward
- Your agent reads notes but doesn't truly remember
Part of the openclaw-inner-life bundle. Requires: inner-life-core