da-skills-auditlisted

# /da-skills-audit — keep the toolkit from rotting Skills degrade in a specific way: they accumulate. Every installed skill's `description` is resident in context at all times, so the more there are, the less of each one survives — and the model picks by matching a request against exactly those descriptions. A toolkit that grows without pruning stops selecting correctly, and nothing announces it. **This skill never modifies anything.** It reports, and proposes. Removals are yours to approve. **This skill reads files. It does not measure usage, and it is not an eval.** Four tools, and you want more than one: | Tool | Reads | Answers | |---|---|---| | **`/skill-doctor`** | the loaded set, with usage | **which loaded skills are unused and costing context.** Run this *first* — it answers what this skill cannot. | | **`/doctor`** | settings and the listing | the listing's real context cost and its biggest contributors; slow hooks; duplicated instructions | | this skill | the files on disk | over-constraint, overlapping triggers, Cursor incompatibility, oversized bodies, the `AGENTS.md` invariants | | **`anthropic-skills:skill-creator`** | with-skill versus without-skill runs | **whether a skill actually helps.** Aggregates pass rate, time and tokens into `benchmark.json`; measures trigger accuracy with should-fire / should-not-fire prompts. | Open the report by naming which of the four were run. **A clean file-level audit says nothing about whether the skills help** — that ne