Conciseness

The body is dense and operational — a path cheat sheet, a workflow, and anti-patterns with no filler or explanations of concepts Claude already knows. It sits at anchor 4 rather than 5 because the 'graph is an index, not the source' point is stated three times (intro, 'When NOT to use', workflow step 5) and the intro paragraph restates frontmatter content, minor trimming opportunities.

Actionability

Every command is concrete and copy-paste ready (`cat ~/.deeplake/memory/graph/query/<pattern>`, `show/<handle-or-pattern>`, `path/<from>/<to>`), with placeholders explained (handles come from `find/`, multi-token AND via `query/<a>+<b>`) and expected output described for each. This matches the fully-executable anchor 5.

Workflow Clarity

The Workflow section gives a clear decision-based sequence (broad → index.md, symbol → find/query, relationships → show/neighborhood, flow/impact → path/impact, code → Read), and the Anti-patterns section supplies explicit feedback loops for failure modes: 'Incoming (0)' → confirm in source, stale build → read live source, `find/` misses → try multiple keywords. Anchor 5 fits; anchor 4 would require some checkpoint to be merely implicit.

Progressive Disclosure

The skill has no bundle files (references/scripts/assets are absent) and the ~90-line body is well-sectioned with clear headers, so it is scored against that actual structure. It earns anchor 4 — good structure, content appropriately placed inline, easy navigation — but the under-50-line exception for a 5 does not apply, and sections like the cheat sheet and anti-patterns are dense enough that they could split into references if the skill grows.

Specificity

The description names the domain concretely — "Query the local code graph (functions, classes, calls, imports) through the Deeplake mount at memory/graph/" — but offers essentially one action (query); the parenthetical lists graph contents, not additional capabilities. It matches anchor 3 (domain + 1-2 concrete actions, not comprehensive) rather than 4, which expects several distinct specific actions like find/show/impact/path that the body actually supports but the description omits.

Completeness

It explicitly answers both: what ("Query the local code graph (functions, classes, calls, imports) through the Deeplake mount") and when ("Use when the user asks structural questions about the codebase") with concrete quoted trigger phrases. This matches anchor 5; anchor 4 would require the 'when' clause to be less explicit, which it is not. Third-person voice is used correctly.

Trigger Term Quality

Quoted triggers are natural user phrasings — "what calls X?", "what does Y import?", "where is Z defined?", "what is the architecture / which subsystems exist?" — plus "structural questions about the codebase". It falls at anchor 4 (good coverage, a few natural terms missing) rather than 5 because common variations like "who uses this function", "dependencies", "blast radius / impact of a change", or language terms (TypeScript, Python) are absent.

Distinctiveness Conflict Risk

A clear niche — AST-derived code-structure queries against a Deeplake mount — with trigger phrases unlikely to collide with grep/read or general coding skills. The specific mount path and relational-question framing make conflict risk minimal, matching anchor 5.