Summary

Many skills/legal/ SKILL.md files shipped with trailing editorial notes from a prior rewrite process — sections like **Key changes from the original:** with bullets describing token reductions, structural changes, etc. A handful also contained leaked LLM permission-prompt chatter ("Could you grant write permission so I can save the file?"). None of this belongs in published skill content.

This PR truncates the trailing pollution block from 305 SKILL.md files. Only the trailing editorial/chatter section is removed — no legitimate skill content is touched.

The reference for what "clean" looks like is skills/legal/real-estate-transaction-summary/SKILL.md, which the user pointed to as the desired end state.

Detection

Pollution shapes encountered:

1. --- separator → **Key changes from the original:** heading + bullets (most common)
2. --- separator → editorial heading without bold markers (e.g. Key changes from the original:)
3. --- separator → **Changes from original:** (no "Key")
4. --- separator → leaked LLM chatter ("It looks like I don't have write permission…")
5. Editorial line + bullets, **with no preceding --- separator` (4 files)

Verification

After cleanup, the following greps return no hits across skills/legal/:

• Key changes from (the )?original
• ^(\*\*)?(Key changes|Key improvements|Changes from)
• ~[0-9]+ lines → ~?[0-9]+ lines / cutting token cost
• write permission|grant write|approve the file write|save the rewritten|copy the (content|output) above

The cleanup script is idempotent (a second run modifies nothing).

Test plan

• Spot-check a handful of files to confirm no real content was removed
• Confirm the verification greps stay empty after merge

Files changed (305)

🤖 Generated with Claude Code