Fetch GitHub profiles for given usernames and score each against required skills. Returns structured candidate data for technical recruitment ranking.
Single tool with clear naming (rank_candidates) and reasonable description (156 chars). Input schema is present with types (list[str], string) and parameter descriptions. However, output schema is undocumented, the tool returns a JSON string but the structure is only visible by reading the implementation code. Error handling exists (rate limiting, 404, API errors) but lacks recovery guidance. No tool annotations (readOnlyHint, etc.). Parameter 'skills' uses free-form string instead of structured input, inviting format errors.
Fetch GitHub profiles for the given usernames and score each against the required skills. Returns structured candidate data as a JSON string — pass this output directly into the ranking_report prompt.
Output schema undocumented. Tool returns JSON string with nested structure (ranking array, error fields, metadata) but LLM cannot see the schema. Forces LLM to parse unstructured output or guess field names.
Parameter 'skills' accepts free-form comma-separated string instead of structured array. Invites format errors (extra spaces, typos, case mismatches). Should be array[string] with enum or pattern constraint.
Error responses lack recovery guidance. Returns JSON with error flags (rate_limited, api_errors) but does not tell LLM what to do next (retry later, check token, try different usernames).
No tool annotations. Tool is read-only (fetches GitHub data, no mutations) but lacks readOnlyHint annotation. Prevents clients from optimizing caching or parallelization.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Prompt 'ranking_report' is present but tool description does not mention it. LLM may not discover that output should be piped into this prompt for final formatting.