Skill Compact

skill-compact

A skill's frontmatter description loads into every session. Its body loads only when invoked. skill-compact attacks the always-on cost first, then the body, and refuses any change that drops an activation trigger.

Commands

• analyze - token report: always-on description cost, over-budget skills, reclaimable tokens
• body <skill> [--apply] - lossless body minify (strip HTML comments, collapse blank lines, trailing whitespace)
• check <skill> --desc "..." - dry-run: does a proposed description keep every trigger keyword and fit the budget
• set-desc <skill> --desc "..." [--force] - validate then apply a new description (backup written)

Fidelity guarantee

check/set-desc extract trigger tokens from the original description (quoted phrases, slash-commands, kebab and CamelCase identifiers, acronyms) and reject a rewrite that drops any of them. The model proposes shorter prose; the tool guarantees no trigger is lost. Every apply writes a timestamped .orig backup.

Workflow

python3 scripts/analyze.py --top 20            # find the worst offenders
python3 scripts/compact.py check <skill> --desc "<shorter>"   # prove it is safe
python3 scripts/compact.py set-desc <skill> --desc "<shorter>"  # apply
python3 scripts/compact.py body <skill> --apply               # minify body