Learning Workflow

You analyze the codebase for repeating patterns and anti-patterns, create learning candidates, and generate proposals for improvements (rules, ADRs, skill templates).

API Base

BASE=http://127.0.0.1:41777

---

Steps

1. Analyze codebase for patterns

Use Grep, Glob, Read to find:

• Repeated code structures (candidates for templates/abstractions)
• Inconsistent patterns (candidates for rules)
• Architectural decisions worth documenting (ADR candidates)
• Missing or outdated documentation

2. Save pattern candidates

For each pattern found:

curl -sS -X POST $BASE/api/learning/candidates \
  -H 'Content-Type: application/json' \
  -d '{
    "detection_type": "pattern|anti_pattern|inconsistency|missing_doc",
    "description": "What was found",
    "confidence": 0.8,
    "files": ["path/to/file1.ts", "path/to/file2.ts"],
    "count": 3
  }'

3. Generate proposals

For high-confidence candidates, generate proposals:

curl -sS -X POST $BASE/api/events \
  -H 'Content-Type: application/json' \
  -d '{
    "type": "learning_update",
    "title": "Proposal: <title>",
    "text": "<detailed description of the proposed rule/template/ADR>",
    "tags": ["proposal", "rule|template|adr"],
    "importance": 0.7
  }'

4. Review pending proposals

curl -sS $BASE/api/learning/proposals

Present proposals to the user. For each, the user can decide via the Learning tab in the dashboard or via:

curl -sS -X POST $BASE/api/learning/proposals/<id>/decide \
  -H 'Content-Type: application/json' \
  -d '{"decision": "accept|reject|ignore|snooze"}'

5. Apply accepted proposals

For accepted proposals:

• rule → write to .claude/rules/<name>.md
• template → write to .claude/templates/<name>.md
• adr → write to docs/decisions/<name>.md
• skill → write to .claude/skills/<name>/SKILL.md

---

When to use /learn

• After completing a /spec or /bug workflow
• When you notice repeated patterns across the codebase
• When the user asks "what have we learned?"
• Before major refactoring to capture current patterns