Execute Cypher or natural language query against the code graph

Rebuild code graph from source

Set active workspace/project context for subsequent graph tools

Report graph/index/vector health and freshness status

Summarize temporal graph changes since txId, timestamp, git commit, or agentId

Explain code element with dependency context

Find architectural patterns or violations in code. Requires pattern (a search string describing what to find, e.g. 'circular dependencies', 'unused files', 'layer violation'). Optional type selects the detection mode: 'circular' = circular dependency detection, 'unused' = files with no relationships, 'violation' = architecture layer rule violations, 'pattern' = general semantic pattern search.

List all MCP tools and their availability in the current session, grouped by category

Normalize and validate tool argument contracts before execution

Search code semantically using vector similarity

Find code elements similar to a given function or class by vector similarity. Requires elementId — use the id field returned by graph_query or code_explain (not a symbol name or natural language string). Optionally set threshold (0–1, default 0.7) and limit. Returns similar elements with names and file paths.

Cluster code elements by directory proximity and vector similarity. Requires type (function | class | file). Returns clusters with member counts and samples — useful for understanding module boundaries and finding groups of related code. Depends on Qdrant embeddings.

Compare graph-stored metadata properties between two code elements. Requires elementId1 and elementId2 — use the id fields from graph_query or code_explain results (not symbol names). Returns changed property keys and left/right-only properties. Note: compares graph metadata, not source-code semantics or embedding similarity.

Suggest test cases for a code element. Requires elementId — use the id field returned by graph_query or code_explain (not a symbol name). Returns suggested test names, types, and coverage gaps based on the element's structure and similar existing tests.

Build a single-call task briefing using PPR-ranked retrieval across code, decisions, learnings, and blockers

Return relevant exact source lines with optional dependency and memory context

One-shot project initialization: sets workspace context, triggers graph rebuild, and generates .github/copilot-instructions.md if not present. Use this as the first step when onboarding a new project or starting a fresh session.

Analyze a repository and generate two files: a lean .github/copilot-instructions.md (project-specific facts only — stack, commands, lxDIG bootstrap) and a .github/lxdig-agent-guide.md (full tool-reference guide with correct signatures, decision table, pitfalls, and usage patterns). The guide is read on demand so it does not saturate the ambient instruction context.

Check code files against architecture layer rules. Returns a violations list and statistics. Call with no files to validate the full project, or pass a list of file paths to scope validation. Violations are returned as warnings by default; set strict=true to elevate to errors.

Suggest the best file path and layer for a new code element. Requires name (the identifier, e.g. 'UserService') and type (one of: component, hook, service, context, utility, engine, class, module). Optionally pass dependencies (list of imports the new element will use). Returns recommended path, layer, and rationale.

Discover and index all markdown documentation files (README, ADRs, guides, CHANGELOG, ARCHITECTURE) under the workspace root into DOCUMENT and SECTION graph nodes. Supports incremental mode (skips unchanged files). Emits DOC_DESCRIBES edges linking sections to the code symbols they mention.

Search indexed documentation sections by full-text query or by code symbol name. Returns matching SECTION nodes with heading, source document, kind (readme/adr/guide/…), line number, relevance score, and a short content excerpt. Run index_docs first to populate the index.

Query a reference repository on the same machine for architecture insights, design patterns, conventions, or code examples. Useful for borrowing context from a well-structured sibling repo when working on the current workspace.

Select tests affected by changed files

Categorize tests by type

Analyze impact of changes

Execute test suite

List tasks or features by status. Pass a query string (e.g. 'all tasks' or 'feature auth') and optionally filter by status (all | active | blocked | completed). Returns matching task or feature nodes with their current status, assignee, and due date.

Update the status of a tracked task. Requires taskId (from progress_query results) and status (new status string, e.g. completed, blocked, in-progress). Optional: notes (progress notes or blockers), assignee, dueDate. Use after completing or blocking a task to keep delivery state current.

Get feature implementation status

Find blocking issues

Persist a structured episode in long-term agent memory. Required: type (one of: OBSERVATION, DECISION, EDIT, TEST_RESULT, ERROR, REFLECTION, LEARNING) and content (the episode text). IMPORTANT: DECISION type also requires metadata: { rationale: '...' } — omitting it returns an error. Optional: entities (related file/symbol names), taskId, outcome (success | failure | partial), sensitive (exclude from default recalls).

Recall episodes by semantic, temporal, and entity relevance

Query decision episodes for a target topic

Synthesize reflections and learning nodes from recent episodes

Claim a file, function, task, or feature for exclusive editing. Conflict detection prevents two agents from claiming the same target simultaneously. Requires targetId (file path or task ID) and intent (natural language description of what you plan to do). Returns a claimId — save it for the matching agent_release call. claimType: task | file | function | feature.

Release an active claim

Get active claims and recent episodes for an agent

Fleet-wide claim view including active claims, stale claims, and conflicts