Loading ToolBench
AI agent for deep financial research on Japanese listed companies. Powered by EDINET DB.
Two tools with partial schema definitions. 'cron' has a complex discriminated union schedule parameter with good structure but lacks output schema documentation. 'heartbeat' has minimal schema detail. Both tools have descriptions (10-20 words), meeting the minimum bar but falling short of LLM-optimized length (50-200 chars). No parameter descriptions visible for most fields. No error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Both tools are WRITE-risk but lack confirmation/dry-run patterns.
Create, list, update, remove, or run scheduled jobs.
View or update the heartbeat checklist (.dexter/HEARTBEAT.md) that controls periodic monitoring.
Missing output schema documentation. Neither tool documents what fields are returned or their types. LLMs cannot plan downstream calls or extract required data.
Descriptions too brief (10-20 words). Rubric baseline is 50-200 chars for LLM-optimized descriptions. Current descriptions lack context on WHEN to use each tool and what it returns.
Parameter descriptions missing or incomplete. 'schedule' parameter uses a complex discriminated union but lacks inline guidance on which discriminant to use. 'action' enum values are listed but not explained (what does 'run' do vs 'list'?).
No error handling guidance. Both tools are WRITE-risk (modify state) but provide no recovery hints. If cron job creation fails, what should the agent do next?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Missing tool annotations. Both WRITE-risk tools lack destructiveHint or idempotentHint annotations. Agents cannot determine if retrying is safe.