Server has 4 tools with basic schemas and descriptions, but significant quality gaps. Tool naming lacks consistency (view-people vs getEmployeeInfo vs calculatePayroll). Descriptions are present but generic (10-50 chars, below 194-char baseline). Parameters lack depth: no enums for constrained values, no format specifications, no range constraints. Error handling is minimal, generic failure messages provide no recovery guidance. Credentials exposed as parameters (authId, authKey) violates secret-injection pattern. Output schemas undocumented. No tool annotations. Composition issues: view-people and view-all-people overlap; getEmployeeInfo duplicates view-people functionality.
Calculate payroll for an employee
Get information about an employee
View all people with optional filtering and pagination
View information about a specific person
Credentials exposed as tool parameters (authId, authKey, partnerKey, clientKey). Secrets in parameters leak into agent logs and traces.
Inconsistent tool naming: verb_noun (view-people, view-all-people) mixed with camelCase (getEmployeeInfo, calculatePayroll). LLMs struggle with inconsistent conventions.
Duplicate/overlapping functionality: view-people and getEmployeeInfo both fetch employee data; view-all-people and getEmployeeInfo serve similar purposes. Violates single-responsibility principle.
Error handling provides no recovery guidance. Generic 'Failed to fetch X' messages give LLMs no actionable next steps. No error categorization (retryable vs fatal).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
Output schemas undocumented. Responses are JSON strings without declared field structure. LLMs cannot plan downstream calls or extract specific fields reliably.