A feed aggregation and processing system with RSS feed management, content enrichment via LLMs, semantic search, and MCP server integration
Scoring was not performed
Three tools (query_app_config_schema, query_app_config, query_rsshub_categories) have empty input schemas with no documented output format. Source code shows {} input but provides no schema, return type documentation, or field descriptions.
Descriptions lack recovery hints and error guidance. Example: 'Query the app config json schema' (23 chars) is too vague. Does it return the schema itself? A reference URL? Supported field types? LLMs cannot infer what to do with the response.
The 'query' tool description includes self-deprecating language ('results may not be accurate, you need to make a secondary judgment') that undermines agent confidence. Rather than warning agents away, improve semantic search quality or document limitations separately.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 34 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 58 | - | v1 |
Parameter 'past' in query tool uses format constraint '^([0-9]+(s|m|h))+$' but this is a regex pattern not visible in schema, only described in text. LLMs cannot parse regex from descriptions reliably. Should use explicit enum or provide valid examples as test cases.
No output schemas documented for any tool. LLMs need to know what fields to expect (e.g., does 'query' return items with id, title, url, published_date? Is there pagination?). Undocumented outputs force agents to guess and risk parsing failures.
apply_app_config is a write/destructive operation but lacks a dry-run, preview, or confirmation step. The description says 'request user confirm the diff' but there is no confirmation_request mechanism in the tool itself. Agents could apply invalid configs without oversight.
Error handling strategy not visible in tool definitions. No error classification (retryable, user-fixable, fatal), no recovery guidance. If a tool fails, agents have no guidance on what to do next.