| name | kg-code |
| skill_type | skill |
| description | Code intelligence over the ingested code graph — cited how/usage/impact answers, cross- repo usages, call graphs, similar code, blast radius, metrics, arch reports — plus templated symbol navigation. Use to understand code via the KG instead of grepping — "how does X work", "who calls this", "impact of changing this", "find the definition". |
| license | MIT |
| tags | ["graph-os","code"] |
| tier | core |
| wraps | ["graph_code","graph_code_nav"] |
| metadata | {"author":"Genius","version":"0.1.0"} |
kg-code
Condensed intent-surface note (Seam 8). Under the default intent surface (MCP_TOOL_MODE=intent), graph_code, graph_code_nav are held back from the default tool list (nothing removed — REST + _execute_tool still reach them exactly as documented below). Two ways to use this skill unchanged: (1) load_tools(tools=["graph_code"]) once per session (as below), then proceed exactly as documented; or (2) call the ask intent verb with the same natural-language request — the resolver routes to graph_code for you and returns the result plus a routing justification. Set MCP_TOOL_MODE=condensed/verbose/both to expose the granular tools eagerly instead.
This skill fronts two code verbs:
graph_code (action): code_context (cited how|usage|impact via target), cross_repo_usages, call_graph (callees|callers|inherits), similar_code, routes, change_coupling, code_evolution, blast_radius, code_metrics, arch_report, adr.graph_code_nav (action): find_definition, find_references, trace_call_graph, impact_of_change, connects (shortest path between two symbols). Start from a symbol or exact node_id; optionally scope by source_system.
Per CLAUDE.md, query the KG before grepping: this is the free native path for "how does this code work?".
Invoke
- MCP:
load_tools(tools=["graph_code"]) (or graph_code_nav), then call it.
- REST twin:
POST /graph/code · POST /graph/code-nav with a JSON body.
Example
graph_code(action="code_context", query="how does the ingestion lane claim tasks", target="how")
graph_code_nav(action="find_references", symbol="reach_user")
Delegation
If graph-os is reachable, offload composite multi-step work via graph_orchestrate (execute_agent / execute_workflow) instead of hand-running the steps — let the local LLM + Loop engine do it, and resolve only the exceptions.
