Conciseness

The body is dense with genuinely non-obvious, project-specific rules (folder contract, gbrain commands, gates, retention posture) and teaches nothing Claude already knows, but it deliberately states key rules multiple times: the Contract section restates the fact-check gate ("At levels `+` and up, no compendium ships without the fact-check claims gate passing"), Anti-Patterns restate the Self-Contained Rule ("Map of pointers ... THE cardinal failure") and retention policy ("Verbatim-archiving sensitive/personal sources when the retention posture says minimize"), and the scope-honesty paragraph acknowledges this duplication ("this section exists for the conformance test"). This fits the 3 anchor (mostly efficient but could be tightened): not 2 because there is no generic padding or explanation of known concepts, not 4 because the cross-section repetition is more than minor.

Actionability

Guidance is fully executable: exact paths ("research/<topic-slug>/sources/NN-<source-slug>"), copy-paste commands ("gbrain lsd \"<the driving question>\" --save --max-cost 5", "gbrain files upload-raw <file> --page research/<topic-slug>/sources/NN-<source-slug>", "gbrain write-request <request_id>", "gbrain pricing set <model> --input <usd-per-1M> --output <usd-per-1M>"), exact frontmatter field names ("untrusted_directives: true", "load_bearing_idea", "tension"), and numeric specifics ("150-300 words", "15-30 quality sources", "every dimension ≥ 7"). This matches the 5 anchor for an instruction-only skill — concrete, specific guidance covering the common cases.

Workflow Clarity

The 4-phase pipeline (Find everything → Archive → Summarize 1:1 → Synthesize) is clearly sequenced with explicit, mechanical validation checkpoints and feedback loops: the claims ledger plus fact-check gate ("run the fact-check skill over the compendium + ledger before shipping"), cold-read validation with a fix loop ("The first 'no' is a skipped precondition — go back and teach it inline"), the cross-modal eval gate ("Ship only if every dimension ≥ 7"), back-link validation at close-out, and error recovery in "When it fails" ("re-read and merge, or poll gbrain write-request <request_id>"). The batch-acquisition risk is covered by validation (test-before-bulk convention, gates), so the destructive/batch cap does not apply; this matches the 5 anchor including the Anti-Patterns checklist.

Progressive Disclosure

Every external reference is one level deep and signaled with its purpose ("see conventions/quality.md for citation rules, quote fidelity, and back-link enforcement"; "see conventions/agent-operator-protocol.md for any gbrain error `code`, exit code..."); the bundle has no reference files of its own, and sections are well-organized with a folder-contract diagram, a depth table, and named phases. This fits the 4 anchor rather than 5 because the ~440-line single file inlines substantial content that an overview-plus-reference split would offload (the extended Low-Bar/High-Ceiling treatise, the Dedup matrix, the Contract restatement); not 3 because nothing is buried or unexplained — all references are clearly labeled with what they contain.

Specificity

The description lists multiple concrete actions with comprehensive coverage of the skill's scope: "archive every primary source verbatim (gated by the user's privacy/retention posture)", "write one 1:1 summary per source", "synthesize a single self-contained compendium page", plus an enumerated depth dial ("base synthesis → grounded primaries → books + counter-canon → saturation"). It matches the 5 anchor (multiple specific concrete actions, comprehensive); it is not a 4 because there is no meaningful coverage gap in the actions named.

Completeness

The "what" is fully explicit and concrete (archive verbatim, 1:1 summaries, synthesize one self-contained page, depth levels), but the description field contains no "Use when..." clause or equivalent affirmative trigger guidance — its only routing is negative ("Distinct from data-research (structured trackers) and perplexity-research (web deltas)"). Per the guideline that a missing 'Use when...' clause caps completeness at 3, it stays at 3: above 2 because the what is clear, but not 4 because the explicit when-guidance lives in `when_to_use`/`triggers` rather than in the description text.

Trigger Term Quality

The description text carries strong natural keywords — "compendium", "deep-research a topic end to end", "synthesize", "prose knowledge synthesis" — but common user phrasings like "definitive guide", "research everything about", and "read them all and summarize" appear only in the separate `triggers`/`when_to_use` frontmatter fields, not in the description field itself. This fits the 4 anchor (good keyword coverage, a few natural terms missing) rather than 5 (comprehensive synonyms inside the description) or 3 (the core terms are exactly what a user would say, not jargon).

Distinctiveness Conflict Risk

It carves out a clear niche and names the neighboring skills it must not collide with: "Distinct from data-research (structured trackers) and perplexity-research (web deltas): this produces prose knowledge synthesis backed by an archived source corpus." This matches the 5 anchor (clear niche, distinct triggers, minimal conflict risk); no neighboring anchor fits better since the differentiators are explicit rather than merely implied.