codebase-knowledge-graphlisted

# Codebase Knowledge Graph — graphify integration Maintain a live map of the codebase — modules, symbols, dependencies, call sites — so change propagation operates on facts, not memory. This is the `graphify` link in the propagation chain. Investigate before asserting impact: query the graph, don't guess what depends on what. ## Fits in the pipeline - **Stage 9 + continuous** — the `graphify` step in `change-propagation`'s chain (PRD → blueprint → roadmap → **graphify** → code). Refresh it before in-code changes so the impact/call-site set is accurate. ## On invocation — detect, then choose a path ### 1. Detect Check whether graphify is installed (try both — on Windows, pip often installs it OFF the PATH): ``` graphify --version python -m graphify --version ``` - If either responds → **resolve & persist the invocation (step 1b)**, then go to "Using graphify". - If not found → present the two options below. Don't silently pick. - Field notes: the pip distribution is named **`graphifyy`** (double-y), so `pip show graphify` reports "not found" even when installed — detect via `--version`, never via pip. And if the user already ran `graphify claude install` (graphify's own Claude Code integration), that's compatible — this skill's flow and the grep-nudge hook push the same direction; don't treat the duplicate CLAUDE.md section as an error. ### 1b. Resolve & persist the invocation (the step that stops agents falling back to grep) The #1 reason a built graph goes unu