DevLens CLI — turn any TypeScript, JavaScript, Python, Go, Rust, or Java codebase into a queryable graph of nodes and typed edges, with per-node functional, technical, and security summaries.
DevLens presents a well-structured tool suite with consistent naming (verb_noun pattern: list_, get_, find_), comprehensive descriptions (avg ~150 chars, well above 34-char baseline), and explicit input schemas with type definitions. All 12 tools are properly registered with descriptions and schemas. However, output schemas are not documented in the visible code, LLMs cannot predict return structures. Parameter descriptions are present but lack actionable constraints (e.g., no min/max for 'limit', no format guidance for 'commitHash'). Error handling is delegated to a generic wrapper (DevLensError) without recovery guidance. Tool composition is strong (single responsibility, proper chaining via graphId/nodeId), but descriptions could better explain WHEN to use each tool vs. alternatives (e.g., find_nodes vs. get_nodes_in_path distinction is unclear).
Search/filter nodes in a graph. Returns COMPACT refs (id, name, type, path, score, 1-line summary) — not source. Combine filters (AND). Use nodeIds for an exact batch fetch. Results are score-ranked and capped by `limit` (default 25).
Impact analysis: the UPSTREAM nodes that depend on the target (who calls/uses it). Answers 'if I change this node, what could break'. Each result carries its hop distance. `radius` defaults to 2; when omitted and the direct (hop-1) fanout is >=100, it returns hop-1 only and sets truncated=true — re-call with an explicit `radius` to traverse deeper, uncapped.
Dependency expansion: the DOWNSTREAM nodes the target calls/uses, out to `radius` hops. Answers 'what does this node depend on'. Same radius/cap behavior as get_blast_radius (default 2, capped at hop-1 fanout >=100 unless radius is explicit).
Full detail for ONE node: metadata, callers (who depends on it), callees (what it calls), and its technical/business/security summaries. Returns precomputed summaries (~50 tokens) instead of source — use this before opening a file. `include` selects which sections; default = everything. `edgeTypes` filters callers/callees by edge kind.
Raw source code for a node. EXPENSIVE in tokens — only call when the summary from get_node is not enough. Returns the exact line range from the analyzed commit.
Output schemas not documented. LLMs cannot predict return structures for any tool, forcing them to reason about unknown fields and risking context loss when chaining calls.
Parameter constraints missing. 'limit' (default 25 - 50) lacks min/max bounds; 'radius' (default 2, capped at hop-1 fanout >=100) lacks clear documentation of cap behavior; 'commitHash' lacks format guidance (SHA-1? short hash?). LLMs will pass invalid values.
Error handling lacks recovery guidance. Generic DevLensError wrapper provides no actionable next steps. E.g., 'graphId not found' should suggest 'Call list_analyzed_repos() to discover available graphs.'
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
List all nodes in a specific FILE or FOLDER. Pass a file path for one file's nodes, or a folder path for everything beneath it (recursive). Great for orienting in an unfamiliar area of the codebase.
High-level orientation for one repo: framework/language fingerprint, route count, and the highest-scoring (most central) nodes. Use this instead of reading package.json or scanning directories.
List nodes flagged with a security concern at or above `minSeverity` (default 'low'), ranked by severity then score. Each entry includes the security summary. Use for security review without scanning the whole codebase.
Return the cohesive cluster (feature/module) that the seed node belongs to: its sibling nodes plus the edges internal to that cluster. Use to understand the bounded context around a node without pulling the whole graph.
Batch-fetch summaries for many nodes at once. Pass the nodeIds (e.g. from a blast-radius result) and get their technical/business/security summaries in one call — far cheaper than reading each file. `include` defaults to all three.
List repositories DevLens has already analyzed. Returns each graphId, repo path, framework, and commit count. Call this first to discover what graphs exist — every other tool needs a graphId. (Also exposed as the devlens://repos resource.)
List groups of nodes that form cyclic dependencies (circular imports/calls). Useful for identifying code smells and refactoring opportunities.
Tool selection ambiguity. find_nodes (search/filter with name, type, path, dir, minScore) vs. get_nodes_in_path (list all in file/folder) distinction is unclear. Descriptions should explain when to use each (e.g., 'Use find_nodes for targeted search; use get_nodes_in_path for exhaustive directory listing').
Pagination not explicit. find_nodes returns 'capped by limit (default 25)' but no total count, next_cursor, or has_more flag. LLMs cannot determine if results are truncated or complete.