Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Madar — Local Code Graph for Coding Agents

    1

    Give your coding agent a local graph of your TypeScript/Node repo so it starts from the right files, routes, and call paths instead of searching everything.

    Free

    2 usesSecurity scanned
    Madar — Local Code Graph for Coding Agents

    Madar — Local Code Graph for Coding Agents

    Example session with this skill installed

    How does password reset work in this repo? Use Madar.

    • Read your context and instructions
    • Compiled the madar

    Summary

    Madar was already installed (0.32.2). I ran it from the repo root, then opened the files it pointed to before answering.

    madar try "how does password reset work?"
    
    [madar try] Built out/graph.json with 34 nodes for a local first proof.
    Graph freshness: fresh
    Indexed files: 8
    Confidence score: 1.00
    
    Recommended first read
    - src/jobs/reset-email-job.ts (enqueueResetEmailJob())
    - src/services/password-reset-service.ts (.requestPasswordReset())
    - src/app.ts (passwordResetService)
    

    Note: Madar wrote its graph to out/. Add out/ to .gitignore if it isn't already ignored.

    Answer

    The account route hands the request to PasswordResetService.requestPasswordReset(). It looks up the user by email, saves a random reset token, enqueues the reset-email job, and records an audit event. It returns { queued: true } even when the email is unknown, so callers can't use it to check which accounts exist.

    Path

    account-routes.ts -> PasswordResetService.requestPasswordReset() -> userRepository.saveResetToken() -> enqueueResetEmailJob()

    Files

    FileWhy it matters
    src/services/password-reset-service.ts:15Token creation, job enqueue, audit event
    src/jobs/reset-email-job.tsSends the reset email
    src/routes/account-routes.ts:8Entry point for the request
    src/app.ts:6Wires the service into the routes

    Confidence

    Madar reported confidence 1.00 on a fresh graph. I opened password-reset-service.ts and confirmed the token save and the unknown-email branch. The path is from static analysis, not a live trace.

    Next steps

    • To let your agent query Madar directly, I can run madar claude install. It writes .mcp.json, a section in CLAUDE.md, and a prompt hook under .claude/. OK to run it?
    • Ask impact on PasswordResetService before changing it, to see what depends on it.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Madar — stop your agent rediscovering your repo

    On a medium or large TypeScript/Node.js codebase, a coding agent usually starts with broad filename searches, rereads the same files, and guesses which route, service, or handler owns the task. Madar gives it a better first move.

    Madar builds a local graph of the repository — files, symbols, imports, calls, routes, handlers, framework roles — and turns each question into a small, task-scoped context pack. The agent starts from likely entrypoints and runtime paths, then reads only the files that matter.

    What this skill teaches your agent

    • Set up Madar safely. Checks Node and the CLI, shows the exact pinned install command, and installs only after you say yes.
    • Get a first answer in one command. madar try "<your question>" builds the graph and returns a recommended first read.
    • Connect your agent. Runs the right installer for Claude Code, Codex, Cursor, Copilot, Gemini CLI, Aider, or OpenCode, and tells you which config files it will touch before it does.
    • Pick the right tool for the question. retrieve for "how does X work", impact for "what breaks if I change X", call_chain for "how does A reach B", pr_impact for "how risky is my diff".
    • Answer honestly. Reports graph freshness and indexing coverage, treats runtime paths as static hypotheses, and opens the named files before explaining or editing.

    Local by design

    • Graph generation runs on your machine, needs no API key, and uploads nothing.
    • .env* files, private keys, and credential stores are excluded from the graph, and the skill tells your agent not to read them.
    • Telemetry is off by default.

    Who it's for

    Developers using AI coding agents on medium-to-large TypeScript or Node.js repositories who ask architecture, runtime-flow, impact, or review questions.

    When it won't help

    Small repos, non-JS/TS codebases, questions about live runtime state, or heavily dynamic code. The skill tells your agent to say so and fall back to normal search.

    Open source (MIT): https://github.com/mohanagy/madar

    TypeScript and JavaScript repositories only. Static analysis: cannot see live runtime state, database values, or heavily dynamic dispatch. Little benefit on small repos. Requires installing the Madar CLI (Node.js 20+); the supported install is a global npm package, run only with your approval.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    2 uses

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Requires

    Node.js 20+
    Madar CLI (@lubab/madar on npm, MIT)
    Node.js

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    File Scopes

    out/**
    .mcp.json
    .claude/**
    CLAUDE.md

    Terminal: runs the madar CLI (try, doctor, status, agent installers). Network: only for the one-time, user-approved `npm install -g @lubab/madar@0.32.2`; graph generation and queries are fully local. Read files: the agent opens the source files Madar points to; .env*, keys and credential stores are off-limits. Write files: Madar writes its graph to out/ and, only when the user approves an agent installer, writes that agent's MCP/rule config (e.g. .mcp.json, CLAUDE.md, .claude/). No source code is modified.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    It teaches your agent to set up and use Madar, which builds a local graph of your TypeScript/Node.js repo and returns a small, task-scoped context pack for each question. Your agent starts from likely entrypoints and call paths, then opens only the files that matter, instead of broad filename searches.

    Madar has project-local installers for Claude Code, Codex CLI, Cursor, GitHub Copilot, Gemini CLI, Aider, and OpenCode. It is built for TypeScript and Node.js repositories and helps most on medium-to-large codebases.

    Node.js 20+ and the Madar CLI. If Madar is missing, your agent shows you the exact pinned install command (npm install -g @lubab/madar@0.32.2) and runs it only after you say yes. Then `madar try "<question>"` builds the graph, and an optional installer connects your agent over MCP. The agent tells you which config files an installer will write before running it.

    No. Madar is static analysis of your source code. Runtime paths it reports are hypotheses from the graph, not live traces, and the skill tells your agent to say so.

    A SKILL.md plus two reference files (MCP tool guide and troubleshooting) that teach your agent how to install, connect, and query Madar safely. Madar itself is a separate, free, MIT-licensed npm package (@lubab/madar) by the same author; the skill does not bundle it.

    Madar builds and queries the graph locally with no API key and uploads nothing. Telemetry is off by default. Your agent may still send prompts and selected file content to its own model provider, depending on that agent's settings.

    New in Code Quality & Review