Deagle-server exposes 6 tools with basic descriptions and partial schemas. Tool names follow verb_noun convention (search, map, sg, rg), which is good. However, descriptions are minimal (10-50 chars), parameter descriptions are sparse or missing, and output schemas are not documented. The 'map' tool lacks clarity on what 'dir' parameter does when optional. Error handling and recovery guidance are absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Schemas present but incomplete, many parameters lack type info or descriptions. This is typical of early-stage code intelligence tools but falls short of production quality.
Health check endpoint
Index a directory into the graph database
Fast regex text search powered by ripgrep
Search for symbols by name in the code graph database
Structural AST pattern search powered by ast-grep
Get graph statistics including node and edge counts
Descriptions are too short (10-50 chars) to guide LLM tool selection. 'Health check endpoint' and 'Get graph statistics' lack context on when to call them or what they return.
Output schemas are not documented. LLMs cannot plan downstream calls or extract fields without knowing what search(), map(), sg(), rg() return. Baseline: 100% of A+ tools document return types.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). 'map' is clearly destructive/write; 'search', 'sg', 'rg', 'stats', 'health' are read-only. Annotations help LLMs reason about side effects and retry safety.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
Parameter descriptions are missing or minimal. 'kind' in search has a description, but 'q' lacks detail on what 'substring match' vs 'fuzzy' means. 'dir' in map/sg/rg lacks guidance on default behavior when omitted.
No error handling or recovery guidance. If search returns no results, or map fails on a directory, LLMs have no actionable next step. Baseline: error responses must tell the LLM what to do next.