Triage an incident evidence bundle and return ranked hypotheses, recommended checks, and escalation-ready notes for a support engineer.
Two tools with complete input schemas and descriptions, but descriptions are minimal (under 100 chars) and lack actionable context for LLM selection. Both tools accept complex nested objects (IncidentBundle, FailedTransaction) with well-structured internal schemas, but parameter descriptions are sparse. No output schema documentation visible. Error handling and recovery guidance absent. Naming is clear (verb_noun pattern), but descriptions do not explain WHEN to use each tool or what distinguishes them.
Classify a failed EDI transaction, rank causes, draft the reply, propose the ticket-two fix.
Analyze an incident evidence bundle and return structured triage output.
Tool descriptions are too brief (under 100 chars) and lack context for LLM selection. Neither description explains WHEN to use the tool, what problem it solves, or how it differs from the other. LLMs cannot infer selection criteria from 'Analyze an incident evidence bundle' alone.
Input parameter descriptions are missing or minimal. The 'bundle' and 'tx' parameters have descriptions, but nested fields (incident_id, summary, timeline, log_lines, etc.) lack guidance on expected format, constraints, or examples. LLMs cannot validate input quality without field-level descriptions.
Output schemas are not documented in tool definitions. TriageToolResult and EdiTriageToolResult are defined in code but not visible in the tool schema. LLMs cannot plan downstream actions or extract fields without knowing what the response contains.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
No error handling or recovery guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. A failed triage call returns nothing actionable to the agent.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Both tools are read-only and idempotent, but this is not declared. LLMs cannot reason about safety or retry behavior without explicit hints.