Conciseness

Every section carries non-obvious, skill-specific guidance — token budgets ("Start at the default --tokens 5000. Retry once at 10000"), latency costs ("deep (4-15s)... deep-reasoning (12-40s)"), and fallback discipline ("Fall through at most once") — with zero padding and no explanation of concepts Claude already knows. Matches the lean/efficient top anchor; there is nothing trimmable without losing operative detail.

Actionability

The full CLI signature with every flag is copy-paste ready, a concrete example invocation is given ("scripts/exa.sh context \"expose a Rust struct to Node via napi-rs threadsafe function\""), and specific numeric defaults (5000/10000/10000 tokens, 600-char highlights, 10 results) cover the common cases. Fully executable guidance matching the top anchor.

Workflow Clarity

The routing decision tree is explicit and fully sequenced: /context first, five enumerated fall-through conditions (empty response, non-code query, recency, provenance, follow-up), narrated fallback with a quoted example, "Fall through at most once" as a stopping rule, and error-recovery loops for empty results ("Remove filters first, then simplify the query") and a missing API key. This is not a destructive or batch operation, so no validation cap applies, and the error-recovery guidance matches the top anchor's feedback-loop standard.

Progressive Disclosure

The body is a lean overview that delegates cleanly: all HTTP lives in scripts/exa.sh (a real bundle file), and references/search-tuning.md (a real 92-line file with no nested references) is signaled with both when-to-read ("when tuning /search filters, choosing a type, or debugging a 400") and when-not-to-read ("Do not read it for a routine /context lookup") guidance. One level deep, clearly signaled, easy to navigate — the top anchor.

Specificity

States the core action concretely ("Search the web and open-source code for library usage, API syntax, and documentation via Exa") and enumerates specific instances such as "verifying an API signature", "checking a config file format", and "resolving an error from a third-party dependency". Falls short of 5 because the capability set is essentially one search operation with use-case variants rather than multiple distinct operations, and stays above 3 because coverage is broad and concrete.

Completeness

Explicitly answers "what" (search web and open-source code for library usage, API syntax, and documentation via Exa) and "when" with two explicit trigger clauses ("Use this whenever verifying an API signature..." and "Also use for questions documentation cannot answer: release timing, breaking changes..."). Both halves are present with concrete trigger phrases, matching the top anchor.

Trigger Term Quality

Includes natural phrases users would actually say: "how a library is used", "config file format", "error from a third-party dependency", plus literal user questions like "is X still maintained", "breaking changes", "security advisories", and "ecosystem comparisons". Coverage spans both code-shaped and non-code triggers with synonyms, matching the comprehensive anchor rather than the "a few natural terms missing" anchor.

Distinctiveness Conflict Risk

The Exa-specific framing ("via Exa", "open-source code", "third-party dependency", "library APIs drift") carves a clear docs/code-verification niche, but "Search the web" as a lead phrase overlaps generic web-search capabilities, leaving minor conflict risk with closely related skills rather than the minimal risk of the top anchor.