Self Improve

self-improve — ChaosEngine learning & adapting

Lean CE-native skill informed by Task Observer methodology (Eoghan Henn / rebelytics, CC BY 4.0 — see LICENSE and UPSTREAM.md). Not a blind clone.

When

• Primary: root-owned Learning Session after confirmed delivery.
• Secondary: explicit operator request mid-session.
• Not: every casual turn — keep always-on cost low.

Dual track

1. Harness — skills, hooks, MemPalace, Graphify, installer/doctor.
2. Product — enhancements for the product under development (queued issues).

Details: references/observation-taxonomy.md. Product track: references/product-track.md (product issues/evals + ChaosGauge link; CLI/doctor/silent-verify over essay tickets). Activation: references/activation.md. Adopt/reject research: references/research-adopt-reject.md. Roadmap: ../../references/self-improve-master-plan.md. Research/explore isolation (harness): ../../references/context-firewall.md.

How (wraps learning.py)

1. Classify each finding (harness vs product; category allow-list).
2. Write minimal fields only: category, title, lesson, proposedChange, benefit, estimatedTokens.
3. Queue through learning.py so privacy gates + GitHub filing invariants hold.
4. Never auto-install skill patches; stage proposals for human/CI review.
5. "Nothing durable" is a valid outcome.

Example:

cat > /tmp/learning-candidate.json <<'EOF'
{
  "category": "tooling",
  "title": "Doctor fix-next missing for Graphify",
  "lesson": "Operators lacked a single repair command after a store probe failed",
  "proposedChange": "Surface repair --component graphify in doctor fix-next",
  "benefit": "Faster store repair on adopter hosts",
  "estimatedTokens": 120
}
EOF
python3 .chaos-engine/learning.py queue \
  --state .chaos-engine-state/learning \
  --upstream Owner/ExampleRepo \
  --candidate /tmp/learning-candidate.json

Metrics / verify (zero-LLM)

python3 .chaos-engine/learning.py metrics
python3 .chaos-engine/silent_verify.py session-start-budget
python3 .chaos-engine/retrieve.py heuristics --top 3
python3 .chaos-engine/significance.py list
python3 .chaos-engine/skill_compress_audit.py audit --skill self-improve
python3 .chaos-engine/meta_optimize.py review
python3 .chaos-engine/draft_skill_pr.py status

SessionStart stays locator-only; prefer CLI over MCP for the same job. Mid-session: mark significant friction only via significance.py (soft fail/deny hooks already write tiny state notes) — never Task Observer. Skill bodies: skill_compress_audit.py proposes compress diffs only; never auto-apply. Periodic: meta_optimize.py review (offline cadence — not continuous). Draft skill PRs: draft_skill_pr.py opt-in only (default OFF; never auto-merge). See meta-optimize and draft-skill-pr.

Local smoke

# From a temp project with ChaosEngine installed, or the source tree:
python3 -c "from pathlib import Path; assert Path('chaos-engine/skills/self-improve/SKILL.md').is_file()"
# Queue one harness + one product candidate (privacy-safe fixtures) via learning.py
# then confirm queue.json grew by two items without secrets/paths.