Graphify

safishamsi/graphify is a Claude Code skill that reads the project's source files and builds a queryable knowledge graph (functions, types, dependencies, cross-file calls). Once installed, Claude Code uses it to answer "where does X get called?" / "what depends on Y?" without having to grep the repo from scratch each time.

Dev-only tooling — not a project dependency, not shipped to end users.

Prerequisites

Python ≥ 3.10 (this project already requires Python for the Whisper transcription module; check with python --version).
Claude Code CLI installed and working.

Install (one-time, per machine)

# PyPI package name is `graphifyy` (double y); the CLI is `graphify`.
pip install graphifyy

# Register the skill with Claude Code (writes ~/.claude/skills/graphify/SKILL.md).
graphify install

# Verify.
graphify --version

On Windows, if graphify is not found after install, add Python's Scripts/ directory to PATH or use pipx install graphifyy instead.

Build the graph for this project

# From the repo root. AST-only extraction — no LLM needed for code files.
# Takes ~30s on this codebase (~300 files, ~2500 nodes, ~5000 edges).
graphify update .

Outputs into graphify-out/ (gitignored — regenerate per machine):

graph.json — the knowledge graph
graph.html — interactive in-browser visualizer (open in any browser)
GRAPH_REPORT.md — readable summary of clusters + key modules
cache/ — per-file extraction cache so subsequent update calls are incremental

Use it from Claude Code

After install, the /graphify slash command is available:

/graphify .                            # rebuild (same as `graphify update .`)
/graphify path "runPhase4" "chunkedGenerate"   # call-chain navigation
/graphify explain src/lib/pipeline.ts          # high-level summary
/graphify query "what depends on RateLimitDialog?"
/graphify affected "completePhase4"            # reverse-traversal — who calls X
/graphify watch .                              # auto-rebuild on file changes

query does a BFS over the graph and caps the response budget; useful for cross-cutting "where in the codebase does Y happen" questions that otherwise need multiple Explore-agent passes.

When to rebuild

graphify update . is incremental — only re-extracts files that changed. Run it:

After a substantial refactor that moved or renamed many symbols
After landing a new top-level feature (e.g. the optional-outputs refactor in commit 92b3fdc or the diagnostic stack in 5552a7d)
If /graphify query starts returning stale results

For continuous live-rebuild during a feature branch:

graphify watch .

leaves it running in a second terminal; the graph stays current with every save.

Why this is dev-only

Graph artefacts (~5 MB on this codebase) are derived data — committing them creates merge noise on every file change.
The skill writes to ~/.claude/skills/graphify/SKILL.md (per-user Claude Code config), not into the repo.
No production code depends on it. Removing it does not affect builds, tests, or runtime.

Uninstall

graphify uninstall            # removes the SKILL.md from ~/.claude/skills/
graphify uninstall --purge    # also deletes graphify-out/
pip uninstall graphifyy