| name | he-learn |
| description | Use for /he:learn; durable guard after repeated misses, review loops, or workflow gaps. |
he-learn
Stage 5/5. Use /he:learn only when he-state.json contains open learning findings after repeated misses, review gaps, or ship findings. Finish the durable guard; do not timebox it.
Read ../workflow-help/references/route-map.md, ../repeated-failure-learning/SKILL.md, and ../skill-creator/SKILL.md before acting.
Contract
- Update
he-state.json before/after each learning step; validate it before loop-complete yes
- If learning findings are empty, do not run this stage;
/he:ship should close with Next: loop complete: yes
- Store learning at the narrow owner: source, script, test, hook, route map, or skill
- Use
skill-creator only for skill/stage-contract changes
- Prefer executable checks; prose-only guidance is last resort
- Failure loop: stay in
he-learn until the durable guard exists and passes
- Exit with the stage receipt: state path, decision, owner/proof, artifacts, blocker, and
Next: loop complete: yes/no. No transcript dump; future runs can resume from state