Codebase indexer for AI agents that extracts and searches code symbols across Go, C, and C++ repositories
Atlas has 6 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-60 chars, mostly short) and input schemas with type definitions. However, descriptions lack depth, they state WHAT but not WHEN or WHY. Parameters are minimally documented. Output schemas are not formally documented; responses are JSON-serialized but structure is inferred from code. Error handling is minimal (basic fmt.Errorf wrapping). No tool annotations (readOnlyHint, destructiveHint, idempotentHint). index_repo and reindex are WRITE operations but lack confirmation patterns or dry-run support. Naming is clear and verb-first (index_, reindex, search, explore, get_, list_), following conventions well. No security issues detected (no secrets in params), but permission gates and audit trails are absent.
Get details about a symbol including its callers and callees.
Get a Mermaid diagram of the repo architecture.
Index a repository for the first time.
List all indexed repositories.
Re-index only files that changed since the last run.
Search for symbols by name, signature, or doc comment.
Descriptions are too brief (10-60 chars) and lack context. They state WHAT but not WHEN to use or prerequisites. E.g., 'Index a repository for the first time' does not explain when to call index_repo vs reindex, or what happens to existing indexes.
Output schemas are not formally documented. Responses are JSON-serialized but structure is inferred from code (e.g., index_repo returns {repo, path, files_indexed, symbols_indexed, duration_ms}). LLMs cannot plan downstream calls without explicit schema documentation.
Destructive operations (index_repo, reindex) lack confirmation or dry-run patterns. Agents can accidentally re-index large repos without warning. No idempotentHint or destructiveHint annotations present.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Error handling is minimal. Errors are wrapped with fmt.Errorf but do not guide recovery. E.g., 'path is required' does not suggest what to do next. No error categorization (retryable vs fatal) or actionable guidance.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot infer which tools are safe to retry or which modify state. search, explore, get_map, list_repos should be marked readOnly; index_repo and reindex should be marked destructive.