Skip to content

ZeroPathAI/zeropath-agent-plugin

Repository files navigation

ZeroPath for Claude Code

CI License: MIT

A security context layer for Claude. Your ZeroPath organization becomes a read/write surface Claude can reason over — findings, cross-repo code search, your team's triage history, custom rules — alongside the file in front of you.

This plugin bundles two pieces of existing ZeroPath tooling for use inside Claude Code:

  • The zeropath-mcp-server for read/write access to your org's findings, rules, and code.
  • The zeropath-cli for local-diff scanning (SAST + secrets + SCA + IaC).

Both are wired in automatically — you only fill in a Token ID and Token Secret at install time.


What Claude can do with the plugin installed

Vanilla Claude can only read what you have open. With this plugin, Claude can also:

  • Search and read code across every repo in your ZeroPath organization — not just what you have cloned locally.
  • See every open, triaged, archived, and patched finding your team has ever worked on, with the reasoning attached.
  • Run security scans against your local edits using your team's existing rule config, custom rules, and suppressions.
  • Ask whether a finding is actually exploitable in your code, not just theoretically reachable.
  • Query your dependency vulnerabilities and CVE database without leaving chat.
  • List the API endpoints your scanners discovered for any repo.
  • Author org-wide custom security rules from a plain-English description, with a dry-run preview.
  • Verify a fix landed by re-scanning and matching fingerprints.
  • Open a PR with the verified diff without dropping back to the terminal.

You don't have to remember new commands. Claude routes to the right capability when your intent matches.


What you actually get

Your appsec team's context, in chat

> is this pattern flagged anywhere in our org?
> why was this finding marked false-positive last quarter?
> what does our team think about this class of bug?

ZeroPath is where your appsec team has already codified what's risky in your codebase - custom rules they wrote after real incidents, suppressions for noisy patterns that don't matter in your stack, severity overrides based on actual exploitability, investigations completed by senior reviewers, the patched/FP/archived history.

Without this plugin, Claude operates without any of that. It can spot a generic eval( but has no idea your team decided three months ago that the one in templates/admin.ts is fine because the input is signed upstream.

With the plugin, every answer Claude gives starts from that institutional knowledge instead of trying to rediscover it.

Cross-repo reach

> have we used this dangerous regex anywhere else?
> show me callers of this function across the org

Claude can search and read code across every repo your ZeroPath org has onboarded — not just what you have cloned locally. Returns file:line citations from repos you've never opened. When you ask about a finding in another repo, it pulls the source from there to ground the explanation in actual code.

A scanner that runs where you write

> is this safe to ship?

/zeropath:scan runs ZeroPath's checks against your local diff in seconds, using your team's existing rule config and suppressions — not a one-off scan. If you try to git push or open a PR with unscanned edits, the plugin gives a soft warning (never blocks).

Verified fixes, not guesses

> fix issue_abc123

/zeropath:fix explains the finding against your real call graph, proposes a unified diff, applies it on your consent, then re-scans to verify the issue is actually gone. Optionally opens a PR for you.

Deep-dive on exploitability

> investigate the criticals

/zeropath:investigate asks ZeroPath whether a finding is actually exploitable in your codebase — not just theoretically reachable.

Custom rules from chat

> flag any direct eval() usage in our checkout service

/zeropath:rule drafts a rule, shows you how many existing files it would flag before you commit, then creates it org-wide on your say-so.


Install

From inside Claude Code:

/plugin marketplace add ZeroPathAI/zeropath-agent-plugin
/plugin install zeropath@ZeroPathAI

Fill in Token ID and Token Secret in the form Claude Code opens. The input is masked and the secret goes to your system keychain. Get a key from Settings → API Keys.

Then fully quit Claude Code (Cmd+Q on macOS) and reopen. The new credentials take effect on the next launch.

Requires uvx on your PATH.


First session — everything auto

You don't run a separate setup or install step. After your token is filled in and you reopen Claude Code:

  • Credentials and CLI are wired up automatically.
  • Your repo is matched against your ZeroPath organization in the background. If it's already onboarded, every read/triage/fix scopes to it without you saying so.
  • On the first session in a fresh project, the banner asks once whether to add a short ZeroPath usage block to your CLAUDE.md so future sessions know when to use which command:
    • /zeropath:claude apply — add it
    • /zeropath:claude decline — never re-ask
    • /zeropath:claude preview — see it first
    • /zeropath:claude remove — strip it later

Slash commands

You don't have to type these — Claude routes to them automatically when your intent matches. ("Is what I just wrote safe?" lands on /zeropath:scan. "What's open here?" lands on /zeropath:findings.)

Command What it does
/zeropath:scan Scan files you edited this session. Drives a fix-and-rescan loop.
/zeropath:findings List open issues, scoped to this repo by default. Also covers "any deps with CVEs?", "what endpoints do we expose?", "how are we doing overall?".
/zeropath:explain Plain-language explanation of a finding — what it is, why it matters in this codebase.
/zeropath:investigate Trigger ZeroPath's AI to assess whether a finding is actually exploitable.
/zeropath:fix End-to-end remediation: explain, propose diff, apply, re-scan to verify, optional PR.
/zeropath:triage Bulk archive / mark FP / update status. Scoped to this repo by default. Always confirms.
/zeropath:rule Author a custom ZeroPath rule from a description. Dry-runs before committing.
/zeropath:claude Apply / decline / preview / remove the ZeroPath usage block in your project CLAUDE.md.
/zeropath:statusline Install or remove the persistent ZeroPath indicator at the bottom of Claude Code. Can extend an existing statusline.
/zeropath:update Check for and apply plugin updates.

Configuration

To change your token later: /pluginzeropathConfigure, then fully quit and reopen Claude Code.

Behavioral toggles (which hooks fire, scan limits, etc.) live in .zeropath/plugin.json (per-repo, checked in) or a per-user file. See the docs page for the full schema.


Status

v1.0.0 — initial public release.

Support

License

MIT — see LICENSE.

About

ZeroPath plugin for AI code agents — bring AppSec scanning (SAST, secrets, SCA, IaC), finding triage, and security context into Claude Code from chat.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors