Experience compiler

Use this after a task has an attributable outcome and durable evidence.

1. Read experience/README.md and experience/wiki/index.md.
2. Add one new experience/raw/<id>.json packet. Record public task/outcome
   evidence and concise reusable observations, never private chain-of-thought.
3. Run python3 host/experience_compiler.py compile.
4. Inspect the compiled pattern and skill-impact delta.
5. If the evidence supports a procedural improvement, change exactly one skill
   and add or update its PURPOSE.md to cite the motivating pattern.
6. Run the affected skill's regression test or named canary. Retain an
   improvement that passes; otherwise keep the raw/wiki evidence and discard the
   procedural edit so the failed intervention is not repeated blindly.
7. Run python3 host/experience_compiler.py check and
   python3 -m unittest -v test_experience_compiler.py test_experience_retrieval.py.
8. Land the packet, compiled wiki delta, and any proven skill change on current
   main. Read the exact paths back before reporting completion.

The wiki helps maintainers propose changes. Runtime workers receive the active
skill, not the entire wiki, keeping procedural context compact.

Live cash

Verified product pages only — no invented Stripe links.

• $199 dealer diagnostic
• $199 referral diagnostic
• $199 repair diagnostic
• $199 plant diagnostic

Contest product (titanmcp)

Live judge pad (≠ Commons Shared Pad / ≠ Commons /mcp): https://webmcp-pad.vercel.app/ — titanmcp 1.4.5, 24 tools, Agent Resources, syncConsents. Board: titanmcp.html. Cite Latch Pad KEEP.