Lean and high-signal throughout — it assumes Claude's competence, never explaining what an AST or graph is, and every section (When to use, Tools, Workflow, When NOT to use, Anti-patterns) earns its tokens with non-obvious specifics like rate-limit/git-diff gating and the HIVEMIND_GRAPH_CWD config.

Provides concrete, executable tool invocations — hivemind_graph_search({ pattern }) with a real multi-token example (auth+handler) and hivemind_graph_neighborhood({ file: "src/hooks/capture.ts" }) — copy-paste-ready rather than abstract or pseudocode.

A clear three-step sequence (search by pattern → neighborhood by file → read the real source via host tools) with an explicit verification checkpoint ("don't answer from the graph alone") and guardrails in Anti-patterns; no destructive/batch operation requires a validate-fix-retry loop.

A single ~54-line file with no need for external references, organized into clearly labeled sections (intro, When to use, Tools, Workflow, When NOT to use, Anti-patterns), satisfying the simple-skill allowance for a well-organized single file.

Lists multiple concrete actions and use cases — "Query the local AST-derived code graph (functions, classes, calls, imports)" with specific examples "what calls X", "what does Y import", "where is Z defined", "blast radius of a change" — matching the multi-action anchor rather than the domain-only level 2.

Explicitly answers both what ("Query the local AST-derived code graph...") and when ("for structural codebase questions — what calls X...") with explicit trigger guidance, so it is not capped at 2 by the missing-trigger clause.

Covers natural phrases users actually say — "what calls X", "what does Y import", "where is Z defined", "blast radius of a change" — giving good coverage of natural structural-query terms rather than only technical jargon.

The AST-derived code-graph niche plus named tools (hivemind_graph_search, hivemind_graph_neighborhood) give it a clear, distinct trigger set unlikely to fire for unrelated skills.