Loading ToolBench
Single tool 'accessreview_scan' has a minimal schema with only one parameter ('target': string) and no type constraints, enums, or validation guidance. Description is generic (58 chars) and lacks actionable context about when/why to use it, what it returns, or error handling. No output schema documented. Parameter description is missing entirely. Tool does not follow verb_noun naming (should be 'scan_access_review' or similar). No error recovery guidance. Fails most production-grade tool patterns.
Periodic user-access-review (UAR) campaign runner. Returns JSON findings.
Parameter 'target' lacks description and validation constraints. LLM cannot infer whether to pass a file path, directory, system name, or URL. No format, range, or example guidance.
No output schema documented. LLM cannot predict return structure (JSON fields, types, pagination). Response is opaque string 'JSON findings', agent must parse blindly.
Tool description (58 chars) is too generic and lacks actionable context. Does not explain WHEN to call it, WHAT it scans for, or HOW to interpret results. Missing prerequisites or dependencies.
No error handling or recovery guidance. Tool may fail on invalid paths, permission errors, or malformed input, but description provides no hints for LLM to self-correct or retry.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | 2026-07-28+ | v2 |
Tool name 'accessreview_scan' does not follow verb_noun convention clearly. 'scan_access_review' or 'scan_user_access' would be more discoverable and self-documenting for LLM tool selection.