Split interactive planning into user-global Patchmill skills

[rochecompaan]

Summary

• Split the interactive planning workflow into four focused user-global Pi skills: planning, upload, labels, and cleanup.
• Added safety-focused contracts for idempotent artifact publication, argv/shell-safe label operations, trusted artifact matching, and revalidated destructive cleanup.
• Documented installation, invocation syntax, authority boundaries, handoffs, and cleanup risk on the Patchmill site.

Validation

• npm test — 1136 pass, 0 fail
• npm run lint — pass
• npm run build — pass
• npm --prefix site run build — pass; generated /using-patchmill/interactive-skills/
• git diff --check — pass
• test -z "$(git status --porcelain)" — pass
• Pi skill load checks, npm package inclusion/exclusion, configured planning-skill resolution — pass

Reviews

• Final Codex re-review passed after accepted safety fixes.
• Final thermo-nuclear re-review passed after removing rehearsal-specific branching.
• Final validation-readiness review passed after running required validation commands.

Closes #129

Patchmill run cost

Stage	Model	Prompt tokens	Output tokens	Estimated cost (USD)
Development environment	gpt-5.5	375,909	2,434	$0.6232
Implementation	gpt-5.5	12,511,296	44,950	$8.6873
Implementation	gpt-5.6-sol	5,873,903	53,156	$6.9542
Implementation	gpt-5.6-terra	4,892,813	131,000	$5.3451
Implementation subtotal		23,278,012	229,106	$20.9866
Total		23,653,921	231,540	$21.6097

Prompt tokens include uncached input, cache reads, and cache writes. Cost is an estimate based on Pi's recorded model pricing and includes parent and subagent sessions.