Conciseness

Dense and competence-assuming — no concept tutorials, real flags only — but there is noticeable duplication: the `max_keywords` default is stated in both the TOML comment and Field notes, the `min_score` higher-is-better direction is explained in Field notes and again in Common pitfalls, and the missing `--keywords` flag is stated in both the Keywords section and pitfalls. Anchor 4 (efficient with minor trimming opportunities) fits better than 5.

Actionability

Fully executable throughout: copy-paste CLI commands with real flags (`--config-json` with a `keywords` JSON block, `--detect-language true`, `xberg embed --preset quality`), jq pipelines, complete TOML config blocks with per-field comments, and a runnable Python snippet. The provider table and cache-warm commands cover the common cases.

Workflow Clarity

A reference-style skill with clearly organized per-feature sections and a useful diagnostic note (feature-gated builds silently ignore `[keywords]` rather than erroring), and the pitfalls section anticipates failure modes. However, there are no explicit verification checkpoints (e.g., confirming expected output shape or that keywords actually landed on the result), so anchor 4 fits better than 5.

Progressive Disclosure

Good structure with clearly signaled one-level references ("See `references/advanced-features.md` for the embeddings pipeline and `references/cli-reference.md` for the `embed` and `cache warm` flag sets"), and the Python/config refs are honestly scoped to the sibling `xberg` skill. But the two `references/*.md` paths have no corresponding files in this bundle (no `references/` directory exists), and some inlined reference material (full config-field annotations, provider table) could live in those files — minor organization gaps keep this at anchor 4.

Specificity

Multiple concrete named actions — "extracting keywords (YAKE/RAKE)", "detecting document language", "generating embeddings for RAG and search" — plus implementation specifics like "`--detect-language`" and "the standalone `embed` command". Coverage is comprehensive for the skill's scope, matching the anchor-5 example of multiple specific concrete actions.

Completeness

Explicitly answers 'when' with concrete trigger phrases ("Use when extracting keywords (YAKE/RAKE) from documents — and, secondarily, when detecting document language or generating embeddings") and 'what' ("Covers the keyword config (and its feature gating), `--detect-language`, and the standalone `embed` command"). The 'when' is more explicit and specific than anchor 4 allows.

Trigger Term Quality

Good natural keywords users would say — "extracting keywords", "embeddings", "RAG", "search", "document language" — but a few natural synonyms are missing (e.g., "key phrases", noun-form "keyword extraction"). Not the comprehensive synonym coverage of anchor 5.

Distinctiveness Conflict Risk

YAKE/RAKE keyword extraction is a clear niche with distinct triggers, but "documents" and "embeddings for RAG and search" carry minor overlap risk with a generic extraction or embeddings skill (a sibling `xberg` extraction skill is referenced in the body). Mostly distinct rather than minimal-conflict.