RAG-only knowledge base toolkit with PDF conversion and vector search capabilities
5 tools with clear verb-noun naming (grep_*, corpus_*, best_character_*). All have descriptions (avg ~80 chars, within baseline). Input schemas present with type info and parameter descriptions. However, output schemas are NOT documented, critical gap for LLM planning. No error handling guidance, no recovery paths. Parameters lack constraints (e.g., max_results 0-200 is stated but not enforced in schema). Tool descriptions are functional but generic; they do not explain WHEN to use each tool vs. alternatives (e.g., grep_search vs grep_search_many). No tool annotations (readOnlyHint present in metadata but not in schema). Composition is reasonable, tools chain (search → lookup), but output fields for chaining IDs are not documented.
Find the best matching character and return file globs for constrained search
Get an overview of the corpus layout, file counts, and agent workflow guidance
Lookup character information from the corpus index by name (Chinese, English, or Japanese)
Search the corpus using ripgrep with fixed-string matching, constrained by file globs and result limits
Execute multiple search queries in parallel against the corpus
Output schemas not documented. LLMs cannot plan downstream calls or extract chaining IDs (e.g., what fields does grep_search return? Does it include file paths for follow-up lookups?).
No error handling guidance. If grep_search returns 0 results, what should the LLM do? Retry with different query? Call grep_search_many? Call corpus_overview first? No recovery paths documented.
Parameter constraints stated in descriptions but not enforced in JSON Schema. max_results '0-200' is text, not a schema constraint. LLMs cannot validate before calling.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Tool descriptions do not explain WHEN to use each tool. grep_search vs grep_search_many distinction is unclear, when should an LLM pick one over the other? No guidance on prerequisites or dependencies.
Tool annotations (readOnlyHint) present in metadata but not in JSON Schema. LLMs rely on schema annotations, not metadata, to understand tool safety.