crew-graph
Build and query a code graph of this repository with graphify, keeping it on the filesystem as graph.json and fresh on commit.
Skill by mbadali25
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/mbadali25-crew-graph
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: productivity
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 26, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
- Runs as
- Python package
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/crew-graph~/.claude/crew/config.json
What's inside
- Skill name
crew-graph- Description
- Build and query a code graph of this repository with graphify, keeping it on the filesystem as graph.json and fresh on commit. Use when the user asks to build or refresh the code graph, asks what calls what or what connects to what, or asks how a subsystem is wired. The graph lives on the filesystem only; there is no vault or notes-app export and no config for one.
- License
- GPL-2.0
- Instructions
- 253 lines in SKILL.md · view on GitHub
Before installing
- Ships executable scripts
- Instructions read files in your home folder (~/.ssh, ~/.aws…)
Files (4)
- SKILL.md
- reconcile.md
- crew_upgrade.py · scripts/
- graph_reconcile.py · scripts/
SKILL.md (first lines)
# Crew Graph
Wraps `graphify`, an external CLI that builds a static code graph and answers
questions against it. This skill detects it, builds keyless, and queries it
instead of loading the raw graph into context.
**The filesystem is the home of the graph, and its only home.**
`graphify-out/graph.json` is one file, greppable, refreshed by a post-commit
hook, and queried without loading it. Everything this skill does stays inside
the repo. It writes nothing to a vault, a notes app, or anywhere else — see
**Why there is no Obsidian export** below, which is a closed question, not a
default someone can turn back on.
## Detect — never auto-install
```
command -v graphify
```
Absent means report it and stop; do not install anything without asking.
Offer:
```
uv tool install graphifyy
```
The package on PyPI is `graphifyy` — double-y — while the CLI it installs is
`graphify`. Every other `graphify*` name on PyPI is unaffiliated. Installing
the wrong one fails silently: `pip`/`uv` succeeds, and the command you get is
not this tool. Say the package name out loud when offering the install; don't
just say "install graphify."
The same install adds a second executable, `graphify-mcp` — see MCP server
below. Confirm with `graphify --version`.
## Build
Check the secrets denylist first, every time:
```
python3 ${CLAUDE_PLUGIN_ROOT}/hooks/scripts/crew_graph_ignore.py --root . --check
```
Exit 0 means graphify will read no secrets-denylisted file. On exit 1, stop:
name the paths it printed and offer `--write`, which adds the missing
patterns to `.graphifyignore`. On exit 2, stop and report its reason. Never
build around either, and never build first and check after — a graph built
over a secret already holds it (**Tainted graph** below). Then:
```
graphify . --no-viz --code-only
```
On native Windows, run it as
`python3 "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/crew_shell.py" run -- "graphify . --no-viz --code-only"`
and quote the `crew-shell:` route line (see `crew-setup/platform.md`). On Git
Bash without `python3`, use `python` or `py -3` with the same arguments.
This is the default, always. Both flags matter and neither is optional:
- **`--code-only`** skips docs, papers, and images. Without it, graphify
errors on any repo that has documentation at all — which is nearly every
real repo, this one included — with `no LLM API key found (N doc/paper/
image file(s) need semantic extraction)`. Code-only needs no API key.
`--code-only` dropped from a command in this skill is a bug, not a
simplification.
- **`--no-viz`** skips the HTML visualization. It's unopenable past roughly
5000 nodes, and an agent reads `graph.json` through the CLI, not the HTML,
so building it is wasted work.
Output lands in `graphify-out/`: `graph.json`, `manifest.json`, `cache/`.
Measured on this repo (106 code files): 1478 nodes, 3170 edges, 1.8 MB.
**Docs, PDFs, and images are opt-in.** Extracting them needs an LLM call, so
before offering to run graphify without `--code-only`, tell the user which
API key env var graphify needs and confirm it's set — don't let them discover
the key requirement from an error after the fact.
## Refresh
Prefer `/crew:graph --refresh` over a raw graphify line. It runs the
denylist check above, then the repo's sanctioned command (`graphify update .`
where `GRAPH_REPORT.md` is tracked beside the graph, which keeps the pair
consistent; the **Build** line where it is not), then proves the report's
`## Summary` counts equal `graph.json`'s `nodes` and `links`. It never
commits: whoever ran it commits the pair. `/crew:graph --status` says in one
line whether the graph is current. **Build** above stays the first build.
## Secrets denylist
graphify reads every file its ignore rules do not exclude, and puts the
symbols it finds into `graph.json`. `crew_graph_ignore.py` builds the repo's
denylist from three sources, in this order:
- **built-in**: `.env`, `.env.*` (but not `.env.example`), `*.pem`, `*.key`,
`*.p12`, `*.pfx`, `id_rsa*`, `id_ed25519*`;
- **`.claude/secrets-denylist`**: optional, tracked, gitignore syntax. This
is where repo-specific paths go (`config/`, `/init.php`). A repo's own
secrets guard can read the same file;
- **`Read(...)` deny rules** in `.claude/settings.json` and
`.claude/settings.local.json`, translated to root-anchored patterns. A rule
outside the repo (`~/…`, `//…`) is skipped by name, and a deny-all `Read`
is `unknown`.
**`.gitignore` never protects a tracked file from graphify.** graphify 0.9.65
(`detect.py`, `_is_scan_ignored`) skips `.gitignore` rules for any path in
`git ls-files --cached`. Only `.graphifyignore` and `--exclude` can exclude
a tracked file. So the check is on the root `.graphifyignore` only. A nested
`.graphifyignore` could re-include a path with `!`, so its presence makes
the answer `unknown`, never covered. The matching is git's own
(`check-ignore --no-index`). Where git and graphify parse an edge pattern
differently, that disagreement is an accepted risk, and the real-graphify
fixture test (`test_graph_ignore_graphify.py`) covers the directory,
anchored, extension and nested forms.
The same check runs in `crew_refresh_check.py`, which will not name a graph
refresh while it fails, and on `/crew:status`'s `graph-ignore` line.
## Tainted graphFrom mbadali25/useful-claude-add-ons, GPL-2.0.
More from mbadali25/useful-claude-add-ons
- 0.603score
crew-qa-standards
mbadali25
Harness, QA-review and environment standards plus a report-only audit (parallel tests, pinned linters CI really runs, gate rules that can fail, QA environment…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
crew-best-practices
mbadali25
Community best practices for working with Claude Code - context management, planning, TDD, hooks, skills, subagents, slash commands and MCP - with what crew…
SkillAFreshness: A★ 2Claude Code +9 - 0.599score
crew-diagrams
mbadali25
Author architecture, process, sequence, ER and data-flow diagrams as Mermaid that pass a measured readability check (no crossing lines, nothing drawn through…
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-debugging
mbadali25
Find the cause of a defect before proposing a fix - read errors, reproduce, check recent changes, gather evidence from this repo's code map and verification…
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-providers
mbadali25
Set up, verify, and invoke external model providers - Codex, Kimi Code or GitHub Copilot for QA review, Gemini for design second opinions.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-change
mbadali25
The change-request template, the ten questions a change board requires, and how each field maps onto ServiceDesk Plus, Jira and a local file.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-context
mbadali25
Manage context exhaustion - warn near the limit, write handoff notes, and resume work after a clear or compact.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-memory
mbadali25
Conventions for durable project memory across repos, including Obsidian vault integration.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-notify
mbadali25
Set up outbound notifications to Microsoft Teams or Telegram - deploy results and "Claude stopped and is waiting" questions - and explain the two-way MCP…
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-cloud
mbadali25
Configure the official AWS and Azure MCP servers safely for a repo.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-docs
mbadali25
Keep CHANGELOG.md, README.md, SECURITY.md, TODO.md, ADRs, generated docs, and the API and feature reference current as work lands.
SkillAFreshness: A★ 2Claude Code +9 - 0.595score
crew-brainstorm
mbadali25
Turn a request into an approved direction before it becomes a spec - one question per message, options with the recommendation first, never implement before…
SkillAFreshness: A★ 2Claude Code +9
Signals
- Stars
- 2
- Forks
- 0
- Open issues
- 17
- Last commit (this folder)
- Oct 8, 2026 (21 days ago)
- Last release
- Jul 28, 2026 (3 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- GPL-2.0
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from mbadali25
- 0.603score
crew-best-practices
mbadali25
Community best practices for working with Claude Code - context management, planning, TDD, hooks, skills, subagents, slash commands and MCP - with what crew…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
crew-qa-standards
mbadali25
Harness, QA-review and environment standards plus a report-only audit (parallel tests, pinned linters CI really runs, gate rules that can fail, QA environment…
SkillAFreshness: A★ 2Claude Code +9 - 0.599score
crew-diagrams
mbadali25
Author architecture, process, sequence, ER and data-flow diagrams as Mermaid that pass a measured readability check (no crossing lines, nothing drawn through…
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-debugging
mbadali25
Find the cause of a defect before proposing a fix - read errors, reproduce, check recent changes, gather evidence from this repo's code map and verification…
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-context
mbadali25
Manage context exhaustion - warn near the limit, write handoff notes, and resume work after a clear or compact.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
crew-change
mbadali25
The change-request template, the ten questions a change board requires, and how each field maps onto ServiceDesk Plus, Jira and a local file.
SkillAFreshness: A★ 2Claude Code +9
Cite this page
crew-graph skill. Agent Mart. https://agentmart.dev/skills/mbadali25-crew-graph (last updated 2026-10-24)