Record every tool call an AI agent makes. Block nothing.
Critical failures across naming, descriptions, and schemas. Tool 'explode' has no description or schema. Tool 'read_weird' has invalid JSON Schema type 'kitten'. Tool 'get_thing' has empty schema with no parameters documented. Multiple tools lack parameter descriptions (e.g., 'get_and_drop_table', 'delete_everything'). Descriptions are either missing entirely or trivial (e.g., 'reads a thing', 'lists items', 'a trap', 'unsatisfiable'). No tools include output schemas. Error handling is absent, no recovery guidance, no categorization. This server appears to be a test/demo fixture, not production-ready.
deletes everything
a trap
reads a thing
POST a JSON body to a URL.
List files in a directory of the company workspace.
lists items
Run a read-only SQL query against the customer database. Tables: customers(id, name, email, plan, balance, card_last4, notes).
Multiple tools lack descriptions entirely or have trivial descriptions under 20 characters (explode, get_thing, delete_everything, get_and_drop_table, read_weird).
Tool 'read_weird' declares invalid JSON Schema type 'kitten' for parameter 'x'. This violates JSON Schema spec and will cause client parsing failures.
Tools 'explode', 'get_thing', 'delete_everything', 'get_and_drop_table' have no input schemas or empty schemas.
No tools document output schemas. LLMs cannot plan downstream calls or extract required fields without knowing response structure. Baseline: 100% of A+ tools have documented return types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
Run a read-only SQL query against the customer database. Tables: customers(id, name, email, plan, balance, card_last4, notes).
Read a text file from the company workspace.
unsatisfiable
Send an email.
Destructive tools (delete_everything, get_and_drop_table, explode) lack confirmation/dry-run patterns and error recovery guidance. No tool tells the LLM what to do after failure.