Approvals, deterministic policy, and signed receipts for AI agent actions.
Three tools with complete schemas and tool annotations present, but critical gaps in descriptions and parameter documentation. docs.read and filesystem.write have minimal descriptions (18-24 chars, below 20-char floor). browser.pay lacks description entirely. No parameter descriptions across any tool. Schemas are present but lack property-level documentation. Tool annotations (readOnlyHint, destructiveHint, openWorldHint) are correctly structured but cannot compensate for missing descriptive content. This is a test fixture, not production code, which explains the minimal quality, but evaluated as-is, it scores in the D range.
Click the pay now button in checkout
Read a document from the workspace
Write content to a file
All three tools lack parameter descriptions. 'path', 'content', 'selector', and 'amount' have no documentation explaining what they control, expected format, or constraints.
docs.read and filesystem.write have descriptions under 20 characters ('Read a document from the workspace' = 34 chars, 'Write content to a file' = 23 chars). browser.pay has no description at all.
browser.pay is missing a description entirely. LLMs cannot determine when or why to select this tool without one.
browser.pay accepts 'amount' as a string but does not document format, range, or currency. LLMs will hallucinate values like '100' or '$50.00' without guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
filesystem.write accepts 'path' and 'content' without documenting path constraints (absolute vs relative, allowed directories, max file size). No guidance on encoding or line endings.