skill-overlap
Flag content-redundant SKILL.md pairs as merge candidates — pairwise body cosine similarity over the skill pack, sorted highest-first. Use when the pack feels bloated or two skills seem to overlap, to find "skills A and B are N% token-similar — consider merging" candidates before a skill-lifecycle archive pass. Proposal only — never merges, edits, or deletes a skill.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 48
- Forks
- 16
01 前言 Front Matter
- name
- skill-overlap
- description
- Flag content-redundant SKILL.md pairs as merge candidates — pairwise body cosine similarity over the skill pack, sorted highest-first. Use when the pack feels bloated or two skills seem to overlap, to find "skills A and B are N% token-similar — consider merging" candidates before a skill-lifecycle archive pass. Proposal only — never merges, edits, or deletes a skill.
02 资源文件 2
- 脚本skill_overlap_test.py7.2 KB
- 脚本skill_overlap.py7.6 KB
03 SKILL.md 正文
/skill-overlap — flag content-redundant SKILL.md pairs (never merge)
The one context-cost technique from the Comet/Perplexity writeup that fak did
not already cover (#3930): skill-to-skill content redundancy.
skill-lifecycle archives dead skills and
skill-score audits over-budget ones — neither measures
whether two live skills say the same thing. With 55+ SKILL.md files (many
*-score scorecards), that overlap is where the pack quietly bloats.
This skill measures it and stops. Max action is to PRINT candidates for a
human to judge — like skill-lifecycle's dry-run archive verdicts, it proposes,
it never merges. Deciding two skills are genuinely one is a semantic call the
detector deliberately does not make; there is no --apply.
The witnessed helper
skill_overlap.py runs a pairwise token-similarity primitive over every
.claude/skills/*/SKILL.md body and emits the pairs above a threshold,
sorted by similarity:
python .claude/skills/skill-overlap/skill_overlap.py [--threshold T] [--top N] [--json]
- Front-matter is stripped first — the shared
name:/description:YAML skeleton is structural, not redundant, so it must not inflate the score. - Similarity is body cosine over token frequency with a small function-word
stoplist — "skills A and B are N% token-similar" in the issue's own words. Same
posture as
fak dup/fak traj(block- and query-level near-dup detection), self-contained so the witness runs with no builtfakbinary. - Default threshold 0.5; drop to
--threshold 0.4to see the long tail. shared_topnames the highest-weight tokens both bodies carry — the why behind each score, so a human can judge the merge without re-reading both.
Exit code is always 0: this is advisory, and "no candidates" is a clean answer.
How to read a candidate
A high score means lexical overlap, not that a merge is correct. Two *-score
scorecards may share vocabulary (debt, scorecard, commit) yet tend
genuinely different surfaces. Before proposing a merge:
- Read both
SKILL.mdbodies. Do they tend the same surface, or just share the scorecard idiom? - If genuinely redundant → open an issue proposing the merge, or route the
weaker one through
skill-lifecycleto archive (reversible), never a raw delete. - If distinct-but-similar → the overlap is a discoverability smell; sharpen
each
description:'s "Use when…" trigger viaskill-scoreso the two stop colliding in recall.
Witness
python .claude/skills/skill-overlap/skill_overlap_test.py — a fixture skill
pack with one deliberately-overlapping pair asserts that pair is surfaced above
the threshold while two unrelated skills are not, plus the front-matter-stripped,
proposal-only, deterministic, and archived-skip invariants. This is the skill's
proof-of-correctness; run it after any change to the detector.
04 同仓库更多 Skills
serve-receipt
On-demand research protocol + deterministic receipt for serving a Hugging Face model on a concrete GPU target — fan out research lanes (model facts/files, engine support matrix, serving recipes, KV/state memory math), compute the tiered feasibility verdict with `fak serve-receipt` (never narrated math), and emit a witnessed launch receipt. Use when asked for an "initial config", "serving receipt", or "how do I serve <model> on <GPUs>".
caveman
Ultra-compressed communication mode. Cuts output tokens 65% (measured) by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.
quality-report
Score the repo quality and emit a report card.
ctx-overlay
Compact the context window and prune the memory store when a session grows too long.
quality-report
Score the repo quality and emit a report card.
ctx-overlay
Compact the context window and prune the memory store when a session grows too long.