code-graph

name	code-graph
description	Use when the user asks a blast-radius, impact, "what calls X", "what breaks if", or dependency-chain question about a linked repo. Queries the pre-built graphify code graph instead of multi-hop grep exploration. NOT for general "explain this code" questions — plain file reading wins there.

Code Graph Queries

Pre-built code graphs (when the user has built one) live under <ARGUS_HOME>/graphs/<repoId>/<scopeKey>/graphify-out/graph.json, with build info in the sibling meta.json. List <ARGUS_HOME>/graphs/ to find them; match the repo by the repoId folder name (derived from the repo's remote, e.g. acme-widget-lib).

Before querying

1. Read meta.json. If status is not ok, say so and stop.
2. Compare meta.json.commit with the repo's HEAD (git rev-parse HEAD). If they differ, still answer, but flag in your findings that the graph is N commits behind (git rev-list --count <meta.commit>..HEAD).
3. If no graph exists for the repo, say so and suggest the user build one from the repo chip in the case header. Do NOT run graphify extract yourself — builds take minutes and are user-triggered.

Query flow (always by node ID, never bare labels)

The graphify CLI is on PATH. Bare names are ambiguous — resolve first:

1. graphify explain "SymbolName" --graph "<graph.json>" — returns the node's full ID (e.g. c_users_..._src_ui_map_map), source file:line, and its connections.
2. Blast radius / who-is-affected: graphify affected "<node-id>" --graph "<graph.json>" --depth 2
3. Dependency chain between two symbols: graphify path "<node-id-A>" "<node-id-B>" --graph "<graph.json>"

Output discipline (mandatory)

• affected on a high-degree node can emit hundreds of lines. Redirect to a file in the case directory and read selectively, or use graphify query "<question>" --graph <path> --budget 2000 for capped traversals. Never ingest an unpaged dump into context.
• Graph edges are syntactic (tree-sitter): dynamic dispatch, dependency injection, and virtual calls can be missing or noisy. VERIFY every edge you cite in findings by reading the source location the graph gives you. The graph tells you where to look; the code is the truth.