Single tool 'seataudit_scan' has a basic description (67 chars) but critical gaps in schema documentation and parameter guidance. Input schema visible in code shows only 'target' parameter with type string and minimal description. No output schema documented. No error handling guidance, no parameter constraints (e.g., path validation, size limits), and no indication of what 'findings' structure contains. Description lacks WHEN to use, prerequisites, or expected output format. Naming is acceptable (verb_noun pattern) but parameter 'target' is ambiguous, does it accept files, directories, or both? No guidance on expected return format or error cases.
SaaS license, seat-usage and shadow-IT auditor. Returns JSON findings.
Output schema completely undocumented. Tool returns JSON findings but LLM has no visibility into structure (fields, types, nesting). Prevents downstream tool chaining and forces LLM to guess at response shape.
Parameter 'target' lacks constraints and format guidance. No indication whether it accepts file paths, directory paths, or both; no validation rules (e.g., must exist, max depth); no examples of valid input.
Description (67 chars) is below baseline (194 chars avg for A+ tools) and lacks critical context: WHAT does 'findings' contain? WHEN should this be called vs other audit tools? What are prerequisites? What does 'shadow-IT auditor' mean operationally?
No error handling guidance. Tool may fail on invalid paths, permission errors, or large directories, but description provides no recovery hints. LLM cannot determine if error is retryable or user-fixable.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
No pagination or result limits documented. If scan returns thousands of findings, response could exhaust context window. No indication of max results or how to handle large scans.