What it is, and what it is not
Graphify reads a project folder and writes a graph of its concepts and how they connect: calls, imports and inheritance across roughly 40 languages, plus links into docs, PDFs and media. Your agent then queries that graph instead of re-reading files. It is a developer tool you run locally; it is not a diagram editor, a notes app, or a hosted analytics product.
The package name: graphifyy, not graphify
The upstream README marks this as important: the official PyPI package is graphifyy, with a double y. Other graphify* packages on PyPI are not affiliated. Only the package name differs; the command you run afterwards is still graphify.
Wrong: pip install graphify · Right: pip install graphifyy
Install the CLI
Commands follow the upstream README at commit a0946f962af3 (fetched 31 Oct 2026). Every platform row reflects upstream documentation; AgentSkillsDir has not run these commands independently.
Needs Python 3.10+ and uv (recommended) or pipx. Pick one install method:
uv tool install graphifyy # recommended
pipx install graphifyy # alternative
pip install graphifyy # may need PATH setupIf graphify is not found after uv tool install, the README says to run uv tool update-shell. On Windows, install uv with winget install astral-sh.uv first.
Register the skill in your assistant
graphify install on its own targets Claude Code by design. For other assistants, use the named command. Excerpt from the README platform table:
Claude Code (macOS / Linux)
then/graphify .graphify installClaude Code (Windows)
then/graphify .graphify install --platform windowsPlain graphify install auto-detects Windows; the flag forces it. Install uv first with winget install astral-sh.uv.
Codex
then$graphify .graphify install --platform codexCodex uses $graphify, not /graphify. For parallel extraction the README asks for multi_agent = true under [features] in ~/.codex/config.toml.
Cursor
then/graphify .graphify cursor installGemini CLI
then/graphify .graphify install --platform geminiAgent Skills (.agents/skills)
then/graphify .graphify install --platform agentsFor frameworks that read ~/.agents/skills. The README lists 20+ further platforms (Copilot CLI, OpenCode, Aider, Kiro and more).
One repository only
Add --project to write the skill into the current repository (for example .claude/skills/graphify/SKILL.md or .agents/skills/graphify/SKILL.md) instead of your user profile:
graphify install --project
graphify install --project --platform codexMinimal flow
From the upstream README; AgentSkillsDir has not run it on a repository yet.
- Install the CLI from the official package: uv tool install graphifyy (or pipx install graphifyy). Needs Python 3.10+.
- Register the skill for your assistant: graphify install, or the platform command from the table above.
- In the assistant, run /graphify . on a repository you are allowed to read (Codex: $graphify .).
- Open graphify-out/graph.html, read GRAPH_REPORT.md, and keep graph.json for later questions.
What you get
Three files under graphify-out/:
graph.html- Interactive graph to open in any browser: click nodes, filter, search.
GRAPH_REPORT.md- Written highlights: key concepts, surprising connections and suggested questions.
graph.json- The full persisted graph, queryable later without re-reading your files.
What stays local
- Code: parsed locally with tree-sitter. Upstream says no LLM is used for this pass and nothing leaves your machine.
- Docs, PDFs and images: sent to your assistant’s model, or to an API backend you configure, for a semantic pass. Upstream lists --code-only to skip them.
- Upstream states no telemetry, but logs graph queries to ~/.cache/graphify-queries.log; set GRAPHIFY_QUERY_LOG_DISABLE=1 to opt out.
Graphify or Obsidian?
They do different jobs. Obsidian is a personal notes vault you browse and write in. Graphify builds a project graph your coding agent queries through /graphify, with code mapped from the AST. If you already live in a vault, skill.md documents an optional --obsidian flag that also writes the graph as an Obsidian vault.
Before you install
- This is third-party code that registers hooks and instructions in your assistant. Read skill.md and the source before you install it.
- Listing is not a security review. AgentSkillsDir has not reviewed this code.
- Upstream also promotes a paid Graphify Cloud; it is separate from the open-source skill and not covered here.
FAQ
What is the Graphify skill?
Graphify Labs’ agent skill and CLI (Graphify-Labs/graphify, Apache-2.0) that maps a project folder into a queryable knowledge graph via /graphify (Codex: $graphify).
Is Graphify a drawing tool?
No. It builds a code and docs knowledge graph for agents to query, not a generic diagram editor or whiteboard.
What PyPI package do I install?
graphifyy, with a double y. Do not run pip install graphify; upstream says other graphify* packages on PyPI are not affiliated. The CLI command is still graphify.
How do I install it for Claude Code, Codex, Cursor or Gemini?
Per the upstream README: graphify install (Claude Code), graphify install --platform codex, graphify cursor install, graphify install --platform gemini. Fetched 31 Oct 2026; not yet run by AgentSkillsDir.
Does code analysis leave my machine?
Upstream says code is parsed locally with tree-sitter, with no LLM for that pass. Docs, PDFs and images go to your assistant’s model, or to an API you configure, for a semantic pass.
What are graph.html, GRAPH_REPORT.md and graph.json?
An interactive browser graph; written highlights with suggested questions; and the full graph file for later queries. All three land in graphify-out/.
Graphify or Obsidian?
Different jobs. Obsidian is a personal notes vault; Graphify builds a project graph your agent can query, and can optionally export an Obsidian vault with --obsidian.
Does AgentSkillsDir host my graph or sell Graphify Cloud?
No. This is a source-linked directory entry. Nothing installs or runs from this page.
Has AgentSkillsDir reviewed the code?
No. Listing is not a security review. Read the source and skill.md before you install.
Are the star and version numbers current?
They are a dated snapshot from the GitHub and PyPI APIs; see the last-checked time on the source card.
Related
- Browse all Claude skills in the directory
- Replica Skill for Claude Code
- Read the official README on GitHub
Corrections and suggestions
Spotted something out of date, or want to suggest a skill? Email a correction. To suggest a resource, submit it for manual review.
