Conciseness

The body is lean and assumes competence ("Prefer `rg` for a simple literal lookup; do not create an index merely because the tool exists"), but the version pin "@colbymchenry/codegraph@1.6.0" also appears in prose outside any old-patterns/deprecated section, and the closing line "upstream savings are not an Arnesto benchmark" is cryptic. Not 5 because of these minor time-sensitive/trimmable elements; not 3 because there is no padded explanation of concepts Claude already knows.

Actionability

Provides copy-paste-ready pinned commands for init/status/query/callers along with the required environment exports, but "run `sync .` with the same version and environment" leaves the sync command to be reconstructed, and no expected `--json` output shape is shown. Not 5 because of these minor gaps; not 3 because the primary commands are fully executable and complete.

Workflow Clarity

Clear scope → CLI → MCP → verify sequence with explicit validation ("Confirm non-zero indexed files/nodes, resolve a known symbol to its real file, and check at least one known caller") and error-recovery loops ("Fall back to source search if indexing fails or the language is unsupported"; "Do not treat an empty or stale index as evidence of no callers"). Not 4 because checkpoints and fallbacks are explicit rather than implicit.

Progressive Disclosure

A well-organized ~65-line overview with a single clearly signaled one-level-deep reference — "See [usage and sources](references/usage.md) for setup and test evidence" — and the referenced file exists in the bundle with no further nesting. Not 4 because navigation is unambiguous and the operational/evidence split is appropriate; the only redundancy (env vars and the init --yes warning repeated in the reference) is minor.

Specificity

Names four concrete actions — "find symbols, callers, dependencies, and change impact" — plus both mechanisms ("local CodeGraph CLI or an already configured MCP server"), matching the comprehensive multiple-actions anchor. Not 4 because the tool's core query surface is fully enumerated rather than having minor coverage gaps.

Completeness

Explicitly answers what ("Use the local CodeGraph CLI or an already configured MCP server to find symbols, callers, dependencies, and change impact in a repository") and when ("Use when the user names CodeGraph or repeated structural code exploration would benefit from a reusable index") with concrete trigger conditions and exclusions. Not 4 because the "Use when" clause is explicit and specific rather than only suggestive.

Trigger Term Quality

Good natural keyword coverage ("symbols", "callers", "dependencies", "change impact", "CodeGraph", "structural code exploration") plus useful negative triggers ("Not for a simple text lookup"). Not 5 because common user phrasings like "find usages", "call graph", or "code navigation" are absent; not 3 because coverage goes well beyond a few relevant keywords.

Distinctiveness Conflict Risk

Clear niche with a named-tool trigger ("the user names CodeGraph") and explicit disambiguation from generic search ("Not for a simple text lookup, prose-only repositories, or automatic installation into every agent"), leaving minimal overlap risk with other skills. Not 4 because both positive and negative boundaries are stated, leaving little room for wrong-skill triggering.