skill-slim

name	skill-slim
description	Audit SKILL.md files for always-loaded token weight and flag what to defer to references/ or scripts/. Use when a skill feels heavy, when trimming context/CLAUDE.md, or before publishing a skill.

Every SKILL.md loads into context whenever its skill is available. Keep it a thin trigger + pointers; push detail to progressive-disclosure references/*.md and executable scripts/.

Run

python3 scripts/analyze.py path/to/SKILL.md [more.md ...]

Reports per file: estimated always-loaded tokens, budget overage, movable inline-code/long-prose weight, existing progressive-disclosure links, and an estimated post-slim size. Sorted worst-first.

Act on the report

• Inline code/examples → move to scripts/ and reference by path; the model runs them, it doesn't need them in context.
• Long prose blocks → if re-derivable by the model, cut; if reference material, move to references/<topic>.md and link.
• Many headings / deep doc → keep the trigger + top pointers in SKILL.md, split the rest into references/.
• Re-run until under budget (~500 tok). Verify the slimmed skill still triggers on the right prompt and its scripts run.

Details, heuristics, and the budget rationale: references/method.md.