← ClaudeAtlas

purge-private-vocablisted

Use when reader-facing text or plans contain plan-local coinages, abbreviations, layer labels, or finding IDs that may be opaque without their source plan.
YasuakiOmokawa/skills · ★ 0 · AI & Automation · score 70
Install: claude install-skill YasuakiOmokawa/skills
Make the target readable without its source plan. Code structure, comment placement, and naming design are out of scope. ## Input and authority Resolve target from an explicit path or inline text, then the current non-delegated session. Resolve the source from an explicit path or a plan/spec identified in the current task; do not assume a particular plans directory. For multiple outputs, count occurrences across the set but define a retained shorthand once per independently read artifact. If the target remains unresolved, return `不足入力: 対象` and stop without searching for a substitute. Tier controls report detail, never write authority. Choose deep first; a plan is at least standard; otherwise choose lite when its condition matches and standard for the remainder: - lite: target ≤300 characters or at most two distinct candidates; concise report. - standard: other PR/Jira descriptions and documents; full report. - deep: design doc, RFC, public material, or 2000+ characters; also apply the deep preparation in [references/execution-contexts.md](references/execution-contexts.md). Edit only when the user or delegating prompt authorized changing the target. Review-only requests return the report. A delegated run without explicit mutation authority does not edit. A plan normally needs no purge, except when its readers lack the upstream analysis that defined `BB-N`, `WB-N`, `IM-N`, or similar labels. Such a plan is never lite: use at least standard and emit the report before edit