AI-first workflow manager for heavy MCP tools with lazy Agent Skills and patchable runs
ToolCapsule exposes 2 tools with basic schemas and descriptions. Both tools have input schemas with required fields and property descriptions, meeting minimum structural requirements. However, descriptions are terse (create-doc: 62 chars, fetch-doc: 44 chars), parameter descriptions lack context about expected formats or constraints, and output schemas are entirely undocumented. Tool annotations (destructiveHint, readOnlyHint) are present but descriptions do not explain when to use each tool or what they return. No error handling guidance, no pagination support, and no composition hints. The mock server in tests/fixtures is a fixture, not production code, limiting confidence in actual implementation quality.
Create a document from Markdown. Use local files for large payloads.
Fetch a document by URL or ID.
Parameter descriptions lack format/constraint guidance. 'markdown' and 'doc_id' have minimal descriptions; no mention of expected length, format, or valid values.
Output schemas completely undocumented. No specification of what create-doc or fetch-doc return, forcing LLMs to guess structure and fields.
Tool descriptions are below baseline (create-doc 62 chars, fetch-doc 44 chars vs. baseline 194 chars). Descriptions do not explain WHEN to use each tool or what distinguishes them.
No error handling guidance. Responses do not indicate what to do if document creation fails, if a doc_id is invalid, or how to recover.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Tool composition unclear. No indication whether fetch-doc returns a doc_id that can be passed to create-doc, or what fields are needed for chaining.