Conciseness

Mostly efficient — every command and error code carries real operational information and no basic concepts are explained — but there is systematic duplication that could be tightened: the "Outputs" section and the Contract's Outputs bullet are near-verbatim, and "When it fails", "Failure modes", and "Anti-Patterns" each restate PACK_READONLY/STILL_REFERENCED/LOCK_BUSY handling across three sections. Inline time-sensitive version trivia ("v0.40.6.0", "pre-v0.40.6 brains silently ignored") is not confined to a deprecated/old-patterns section, which the guidelines penalize. Not a 4 because the redundancy spans multiple full sections rather than minor trimmable spots; not a 2 because nothing is concept over-explanation or padding.

Actionability

Fully executable throughout: copy-paste-ready commands with real flags for every phase (`gbrain schema active --json`, `fork gbrain-base mine`, `add-type researcher --primitive entity --prefix people/researchers/ --extractable --expert`, `lint --with-db`, `sync --apply`), a concrete JSONL batch example, expected output shapes (would_apply counts, sample slugs), and an enumerated error-code list. Matches the anchor-5 standard of covering the common cases with runnable commands.

Workflow Clarity

The 7-phase sequence (Brain → Assess → Propose → Apply → Sync → Verify → Commit) has explicit validation checkpoints exactly where the batch/destructive risk lives: lint gate before mutation, mandatory dry-run before `sync --apply`, and a Verify phase (coverage ≥95% + `whoknows` spot-check). Error recovery is a real feedback loop (retry once on commit_lock_busy, stop and ask on integrate_conflict, never bypass lint). This is the anchor-5 pattern, not 4, because validation is an explicit workflow step rather than an implied one.

Progressive Disclosure

Good structure: clear section headers, one-level-deep references clearly signaled in the Convention blockquotes (`../conventions/brain-first.md`, `../conventions/schema-evolution.md`, `../conventions/agent-operator-protocol.md`), and external docs appropriately linked under Tutorial + vision. No bundle files exist in this skill directory, and nothing is nested two-plus levels. Not a 5 because of minor gaps: the path for schema-evolution.md is inconsistent (`conventions/schema-evolution.md` in the Convention blockquote vs `skills/conventions/schema-evolution.md` in Anti-Patterns), and the ~50-line Output Format JSON block plus the dense git-outcome prose are candidates for a split-out reference file. Not a 3 because references are well signaled and content is genuinely sectioned rather than a wall of text.

Specificity

Lists four concrete actions — "Add page types", "propose new ones from corpus scans", "backfill page.type on existing pages", "audit pack health" — which clearly exceed the 1-2-action anchor. Not a 5 because the skill body's scope also covers link verbs, prefixes, aliases, and expert-routing flags, none of which appear in the description — a minor coverage gap.

Completeness

Explicitly answers both questions: the "what" is the action list (add types, propose from corpus scans, backfill page.type, audit pack health) and the "when" is the concrete clause "Triggers when an agent notices untyped pages, custom domains needing typed entities..., or wants to see what types the pack declares". Matches the anchor-5 example structure (what + explicit 'use when' with concrete trigger phrases) exactly; a 4 would require the 'when' to be less explicit than it is.

Trigger Term Quality

Natural trigger phrasing is comprehensive for this domain: "untyped pages", "custom domains needing typed entities (researcher, contract, deposition)", "wants to see what types the pack declares" — including concrete synonyms (researcher, contract, deposition) that mirror what a user would actually say. It is not a 4 because no obviously common phrasing for this niche is missing; these read as the exact utterances that should route here.

Distinctiveness Conflict Risk

The niche (authoring/evolving the brain's schema pack) is distinct from filing or querying skills, and the body even names those boundaries. Not a 5 because "wants to see what types the pack declares" is a read-only trigger the body itself routes elsewhere (`gbrain schema explain` directly), and phrases like untyped pages sit close to the adjacent brain-taxonomist skill — minor overlap risk with closely related skills.