| name | graphify-ask |
| description | Query the existing Graphify knowledge graph to navigate this codebase fast. Use for "where is X defined", "what calls Y", "how do A and B connect", "explain this module", "what is in graphify". Reads graphify-out/graph.json — no LLM cost. Aliases - graphify-query, ask-graph, code-nav. |
graphify-ask — navigate the codebase via the knowledge graph
The repo graph lives at graphify-out/graph.json (~90K nodes / ~116K edges,
6,800+ communities). Per-directory graphs also exist at <dir>/graphify-out/graph.json.
Querying is free — pure graph traversal, no LLM call.
Interpreter
PY="$(cat graphify-out/.graphify_python)"
If no graph exists
If graphify-out/graph.json is missing, run /graphify-map first.
Three query verbs
explain — one node + its neighbors
"$PY" -m graphify explain "evaluate_kill" --graph graphify-out/graph.json
Returns the node's source location, type, community, degree, and every inbound
/outbound edge (calls, contains, rationale_for, imports, semantically_similar_to).
Best first move when you want to understand one function/class/concept.
query — BFS traversal from a question
"$PY" -m graphify query "what computes the kill threshold" --graph graphify-out/graph.json
"$PY" -m graphify query "how are picks scored" --budget 4000 --graph graphify-out/graph.json
Flags: --dfs (depth-first instead of BFS), --budget N (token cap, default
2000), --context C (filter by edge type, repeatable, e.g. --context call).
Use when you don't know the exact node name — phrase it as a question.
path — shortest path between two nodes
"$PY" -m graphify path "production_scanner.py" "kill_gate.py" --graph graphify-out/graph.json
Shows the chain of edges linking two symbols. "No path found" is a real answer —
it means the two are in disconnected components (genuinely unrelated, or one
isn't wired into the production path — useful for the CLAUDE.md Wire-Up Rule).
Picking the graph
- Cross-cutting question →
graphify-out/graph.json (whole repo).
- Question scoped to one subsystem →
<dir>/graphify-out/graph.json is smaller
and faster (e.g. alpha_engine/graphify-out/graph.json).
Node-name tips
Node ids look like audit_trail_kill_gate_evaluate_kill; labels look like
evaluate_kill() or kill_gate.py. explain matches loosely on the label —
try the bare function name first, then a fuller form if "No node matching" comes
back. If still nothing, the symbol may post-date the graph build → /graphify-refresh.
Reporting
Summarize what the graph showed in plain language and cite real
file:line from the node src/loc fields so the user can jump there. Don't
dump raw NODE/EDGE lines unless asked.
Companion skills
/graphify-map — build the graph/graphify-refresh — update it after code changes
