/learning-evolve

name: learning:evolve
description: "Run pattern detection on session observations to extract and reinforce instincts"
argument-hint: "--project PROJECT_ID"
allowed-tools:
  - Bash

Run the Coco learning system evolution cycle to scan observations and detect new patterns.

cd "$CLAUDE_PROJECT_DIR" && python3 -m coco_learning.cli evolve --project $ARGUMENTS

This command: 1. Loads recent observations from `~/.coco/learning/projects/<id>/observations.jsonl` 2. Applies heuristic pattern matching rules across testing, security, workflow, architecture, debugging, and performance domains 3. Creates new candidate instincts for patterns with sufficient evidence (default: 3+ occurrences) 4. Reinforces existing instincts by incrementing evidence count and recomputing confidence 5. Saves all detected/reinforced instincts to disk as YAML files

After running, display the results and suggest:

• For strong instincts (confidence ≥ 0.7): recommend `/learning:promote <id> --from-project <id>`
• For candidates with low evidence: explain they need more observation data
• If no patterns detected: suggest continuing normal work to accumulate observations