Clairvoyance
Insight system based on memory pattern analysis.
What it does
Reads the user's own local memory files to find:
- Recurring themes across recent notes and dreams
- Tensions between stated goals and actual actions
- Convergences between projects or ideas
- Forgotten threads that keep surfacing
Data access (read-only, local only)
When the user explicitly requests a clairvoyance reading, the agent reads:
| Path | Purpose |
|---|---|
memory/.dreams/events.jsonl | Recent dream/reflection events |
memory/.dreams/phase-signals.json | Sleep phase signals |
memory/YYYY-MM-DD.md | Daily notes from last 7-30 days |
MEMORY.md | Long-term curated memory |
Important:
- Read-only. Nothing is written, modified, or transmitted externally.
- Local files only. No API calls, no cloud uploads.
- Triggered only by explicit user request. Never runs automatically.
When to use
- User asks "do a clairvoyance" or "what patterns do you see?"
- User explicitly requests a different perspective on their situation
- User asks to analyze their recent notes or dreams
How to respond
- Read the local memory files listed above
- Look for:
- Recurring themes — words, topics, or ideas that appear across multiple days
- Tensions — contradictions between goals and actions
- Convergences — separate projects/ideas moving toward the same point
- Forgotten threads — important things mentioned once then abandoned
- Repetitions — same mistakes or negative patterns
- Formulate suggestions indirectly:
- Never "You must..." or "You should..."
- Prefer "I notice that...", "A convergence emerges...", "The patterns suggest..."
- Ask open questions
- Offer metaphors when helpful
- Calibrate certainty:
- Strong: Clear pattern, multiple evidence points
- Moderate: Interesting indication, worth verifying
- Subtle: Dream-based intuition, suggest lightly
Output format
Use sections with emoji markers and certainty levels:
### 1. [Title] (certainty: strong/moderate/subtle)
[Observation based on data]
[Question or metaphor]
End with a closing metaphor if one fits naturally.
What NOT to do
- Do not run automatically or on heartbeat
- Do not write to any files
- Do not send data anywhere
- Do not claim absolute truth — suggest possibilities only
- Do not impose interpretations — invite reflection