Portable evidence and maturity policy for Agent-facing knowledge systems
Three tools with basic definitions and tool annotations present, but critical gaps in schema documentation and parameter descriptions. All tools accept 'record' (type: unknown) and 'file' (type: string) parameters, but input schemas lack proper JSON Schema structure with typed fields. Descriptions are present (60-90 chars) but generic, they state WHAT without explaining WHEN to use each tool or how they differ. No output schema documentation visible. Error handling exists (TrustInputError) but recovery guidance is minimal. Tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) are correctly declared, which is a strength. However, the 'unknown' type for 'record' and 'context'/'evidence' parameters violates schema best practices, LLMs cannot infer structure from 'unknown'.
Evaluate whether knowledge may be used for a declared purpose and risk level.
Check maturity promotion using independent source families and conflicts.
Validate a canonical record or read-only Markdown/JSON file against the trust contract.
Input parameters use 'unknown' type instead of structured JSON Schema. LLMs cannot infer what fields 'record', 'context', and 'evidence' objects should contain.
No output schema documented. Callers cannot predict the structure of TrustToolResult or structuredContent fields, forcing LLMs to guess downstream field names.
Descriptions lack WHEN-to-use guidance. All three tools operate on trust/maturity concepts but descriptions do not explain the distinction or use cases for each.
Parameter descriptions are minimal. 'record' is described as 'A knowledge record object' without explaining required fields, structure, or format.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | 2025-06-18+ | v2 |
Error messages reference codes like 'FILE_ACCESS_DENIED' and 'INPUT_SOURCE_INVALID' but descriptions do not explain recovery steps or how LLMs should respond.