Conciseness

The body is dense with gbrain-specific operational facts and avoids explaining concepts Claude already knows, but the ~60-line opening blockquote, a duplicated visibility sentence ("Every MCP caller, including stdio, and a thin CLI connected to MCP..." appears twice), and an anecdotal "Real example" of the shallow ls undercount could all be tightened. This fits score 3 (mostly efficient but could be tightened) rather than 4, which requires only minor trimming.

Actionability

Guidance is fully executable throughout: copy-paste-ready commands with flags ("gbrain recall --query 'zebra telescope' --budget-tokens 75 --budget-policy query_first --json", "find <dir> -name '*.md' | wc -l", "gbrain sweep --once"), exact error codes, and concrete recovery steps. This matches the score-5 anchor of copy-paste-ready commands covering common cases.

Workflow Clarity

Phases 1–4 are clearly sequenced with numbered steps, validation is explicit ("Confirm authorized writes only after readback"), and error-recovery feedback loops are specified (revision_conflict: re-read, merge, save with new revision; write_pending: poll before claiming the write landed). This matches the score-5 anchor of explicit validation steps with feedback loops.

Progressive Disclosure

References are one level deep and clearly signaled ("See skills/conventions/quality.md for format", "Docs: docs/guides/write-refusals.md"), but none exist in this skill's bundle, and the opening mega-blockquote inlines memory verbs, budget packing, visibility rules, redaction handling, and keyless-brain extraction — content that clearly belongs in separate reference files. This matches score 3 (content that should be separate is inline) rather than 4, where inlining would be only a minor gap.

Specificity

The description lists several specific named actions — "brain-first lookup, read-enrich-write loop, source attribution, ambient enrichment, back-linking" — mirroring the score-4 anchor's structure of several concrete operations with minor gaps. It is not a 5 because the actions are internal domain jargon rather than fully concrete capabilities, and coverage is not comprehensive.

Completeness

The "what" is clear (the core read/write cycle with named operations), but there is no "Use when..." clause; "Read this before any brain interaction" is a circular, weakly explicit when. Per the rubric guideline, a missing explicit trigger clause caps completeness at 3, and score 4 requires a more explicit and specific when.

Trigger Term Quality

Relevant keywords like "brain", "read/write", and "lookup" are present, but phrases such as "ambient enrichment", "back-linking", and "read-enrich-write loop" are not terms users naturally say, and common variations like "remember", "recall", or "save to the brain" are missing. This matches the score-3 anchor of some relevant keywords missing common variations, and is below score 4's good natural-term coverage.

Distinctiveness Conflict Risk

"Brain knowledge base operations" with the trigger "any brain interaction" carves out a clear niche (a personal brain/knowledge-base system) with a distinct trigger, giving minimal conflict risk with other skills — a direct match for the score-5 anchor.