Loading ToolBench
Verifies double-entry ledger integrity and tamper-evidence by checking balance invariants and hash-chained journal entries.
Single tool 'ledgerproof_scan' has a reasonable description (127 chars, within 10-1024 range) but critical gaps in schema completeness and parameter documentation. Input schema exists but lacks parameter descriptions, the 'target' parameter has only a type and description, no format constraints or examples. No output schema is documented, forcing LLMs to guess the response structure. The tool name uses a verb+noun pattern (good), but the description conflates two distinct concerns (integrity checking AND tamper-evidence verification) without clarity on when to use it. No error handling guidance, no pagination support despite potentially scanning large directories, and no indication of what fields the JSON response contains.
Verifies double-entry ledger integrity and tamper-evidence by checking balance invariants and hash-chained journal entries.. Returns JSON findings.
No output schema documented. LLMs cannot infer response structure (fields, types, nesting). Tool description says 'Returns JSON findings' but does not specify what fields are in that JSON.
Parameter 'target' lacks format constraints and validation guidance. No mention of whether it accepts files, directories, or both; no size limits; no error cases (e.g., 'file not found'). LLMs will pass arbitrary paths without understanding constraints.
Description conflates two responsibilities: 'Verifies double-entry ledger integrity AND tamper-evidence by checking balance invariants AND hash-chained journal entries.' Unclear which checks are performed, when to use this tool, or what 'findings' means in context.
No error handling guidance. Tool may fail on invalid paths, permission errors, malformed JSON, or missing files. No recovery hints for LLM (e.g., 'If file not found, check path and retry').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
No pagination or result limiting documented. Scanning a large directory tree could return thousands of findings, exhausting context window. No mention of limits, pagination, or how to handle large result sets.