Loading ToolBench
Business bankruptcy filings — US federal bankruptcy courts.
Single tool 'search_bankruptcy_dockets' has a clear action verb and reasonable description (156 chars), but exhibits critical schema and parameter documentation gaps. Input schema is present with 6 properties, but 4 lack descriptions entirely (_supabaseUrl, _supabaseKey, limit, filedAfter/filedBefore have minimal guidance). No output schema documented. Parameters expose internal implementation details (Supabase credentials as tool params violates secret-injection pattern). The tool description is domain-specific and helpful, but parameter descriptions are sparse or missing, forcing LLMs to infer intent. Error handling returns structured objects (docketUnavailable) but lacks actionable recovery guidance. Baseline: 194 chars avg description (this tool: 156 ✓), 72 chars avg param annotation (this tool: ~30 avg ✗), 100% of A+ tools have param descriptions (this tool: ~33% ✗).
Search business bankruptcy dockets by party name, court, or filing date. Returns case metadata including docket number, judge, and court information. Does not include full filings (RECAP documents); those are served by the court-listener pack.
Supabase credentials (_supabaseUrl, _supabaseKey) exposed as tool parameters. Violates secret-injection pattern, credentials in params leak into agent logs and traces.
4 of 6 input parameters lack descriptions: _supabaseUrl, _supabaseKey, limit, filedAfter, filedBefore. LLMs cannot infer parameter meaning or constraints without explicit text.
No output schema documented. LLMs cannot plan downstream operations or extract required fields (e.g., docket_id for follow-up queries) without knowing response structure.
limit parameter lacks min/max constraints. Unbounded numeric parameters allow LLMs to pass absurd values; schema declares 1-50 but description does not reinforce this.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2025-06-18+ | v2 |
Error response (docketUnavailable) returns structured object but lacks actionable recovery guidance. LLM cannot determine whether to retry, ask user, or escalate.