Loading ToolBench
The server has 3 well-named, action-verb tools with complete input schemas and reasonable descriptions. Naming follows the verb_noun pattern (send_simple_email, send_custom_email, test_smtp_connection) and is clear. All tools have descriptions between 20-30 characters meeting baseline minimums. Input schemas are properly structured with types and descriptions for all parameters. However, output schemas are entirely undocumented, no schema definition for return values, which violates the critical requirement that 'LLMs need to know what fields to expect.' Error handling returns plain strings without structured guidance. The SMTP config parameter in all tools accepts arbitrary dicts without validation constraints, inviting hallucinated fields. Parameter relationships (e.g., mutual exclusivity of 'text' vs 'html' in send_custom_email) are not documented. Missing error classification (retryable vs fatal) and no dry-run/confirmation patterns for destructive operations like send_simple_email.
Send a custom email with full configuration options.
Send a simple email.
Test SMTP connection.
Output schemas completely undocumented. Tools return plain strings but LLMs cannot infer what success/failure responses contain, what fields are available for downstream chaining, or how to parse results.
smtp_config parameter accepts arbitrary Dict[str, Any] with no validation constraints, schema documentation, or enum for known fields (host, port, secure, username, password, from_email). LLMs may pass invalid keys, misspelled fields, or wrong types without validation.
Error responses are bare strings ('Error: ...', 'Error sending email: ...') with no recovery guidance, error classification (retryable vs fatal), or actionable next steps. Violates recovery-guide pattern, agent cannot distinguish between timeout/auth failure/invalid recipient.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 12 | <=2025-11-25 | v2 |
| 2026-03-09 | C | 61 | - | v1 |
send_custom_email accepts 'email' dict with 'text' and 'html' fields but does not document that at least one must be provided, that both are optional, or constraints on attachment MIME types. Parameter relationships are undocumented.
Destructive operations (send_simple_email, send_custom_email) have no dry-run, confirmation, or reversibility pattern. An agent in a retry loop could send duplicate emails without intervention.
Parameter 'is_html' in send_simple_email is boolean but description does not state the default (False). Tool implementation has the default but the schema description should be explicit for LLM clarity.