remember

name	remember
description	Capture behavioral friction as structured observations. Three modes: explicit (user states a lesson), contextual (scan conversation for corrections and patterns), and session mining (bulk extract from transcripts). Stores observations for later synthesis via /rethink. Triggers on: "remember", "lesson learned", "note pattern", "capture friction"

/remember

Capture behavioral friction and lessons as structured observations.

Purpose

The input side of the learning loop. When something goes wrong, when a pattern emerges, or when the user corrects behavior — capture it as a structured observation. Observations accumulate over time and feed into /rethink for synthesis into system improvements. Three capture modes handle different scenarios: explicit (user tells you), contextual (you detect it), and mining (bulk extraction).

Usage

# Explicit — user states a lesson
/remember "Always check for duplicates before inserting into the knowledge base"

# Contextual — scan recent conversation for corrections
/remember --contextual

# Mine session transcripts for patterns
/remember --mine path/to/session-transcript.md

# Mine recent sessions
/remember --mine --recent 7d

# List stored observations
/remember --list

# List observations by category
/remember --list --category process

# Check which categories are ready for synthesis
/remember --escalations

# Tag an observation
/remember "Never send specs to Robert Potter" --category people --severity high

Arguments

Flag	Type	Default	Description
<observation>	positional	—	Explicit observation text
--contextual	flag	false	Scan current conversation for implicit corrections
--mine	path/flag	—	Extract observations from transcript file(s)
--recent	duration	7d	For --mine, look back this far
--list	flag	false	List stored observations
--escalations	flag	false	Show categories ready for /rethink
--category	string	auto	Category: process, people, technical, communication, tool-use
--severity	enum	normal	critical, high, normal, low
--confidence	float	0.5	Initial confidence (increases with repeated observations)
--source	string	auto	Where this observation came from

Workflow

Explicit mode

1. Parse — Extract the core lesson from the observation text.
2. Categorize — Auto-classify into category (or use --category).
3. Deduplicate — Check against existing observations. If similar exists, increase its confidence instead of creating duplicate.
4. Store — Write to ops/observations/{category}/{timestamp}-{slug}.json.
5. Check threshold — If category now has enough observations (default: 5), flag for escalation.

Contextual mode

1. Scan — Read recent conversation history.
2. Detect — Look for correction patterns: "No, I meant...", "That's wrong...", "Actually...", user re-doing work, explicit feedback.
3. Extract — For each detected correction, formulate an observation: what went wrong, what the correct behavior is.
4. Store — Same as explicit mode for each extracted observation.

Mining mode

1. Load — Read transcript file(s) or recent session logs.
2. Pattern match — Identify: repeated mistakes, workflow friction, time wasters, user frustration signals, successful shortcuts.
3. Extract — Formulate observations from each detected pattern.
4. Batch store — Store all observations, deduplicating against existing.
5. Report — Summarize what was mined and how many observations were captured.

Output

Stored observation

{
  "id": "obs-2026-03-20-143500-a1b2",
  "observation": "Always check for duplicates before inserting into the knowledge base",
  "category": "process",
  "severity": "normal",
  "confidence": 0.5,
  "occurrences": 1,
  "source": "explicit/user",
  "created_at": "2026-03-20T14:35:00Z",
  "updated_at": "2026-03-20T14:35:00Z"
}

Escalation check

## Observation Escalations

| Category | Count | Avg Confidence | Ready for /rethink? |
|----------|-------|---------------|---------------------|
| process | 7 | 0.68 | YES (threshold: 5) |
| people | 3 | 0.55 | no (threshold: 5) |
| technical | 12 | 0.74 | YES (threshold: 5) |
| communication | 2 | 0.40 | no (threshold: 5) |
| tool-use | 5 | 0.62 | YES (threshold: 5) |

Run `/rethink "process"` to synthesize process observations into system improvements.

Dependencies

• File system for observation store (ops/observations/)
• Conversation history access (for --contextual mode)
• Session transcript access (for --mine mode)
• /rethink — Downstream consumer of accumulated observations