CLI & agent skills

Put your audit in the terminal — and in your agent's hands.

One binary connects your shell to the same audits and findings you see in the web app. Three bundled skills teach your coding agent to read them, act on them, and start new runs — over the authenticated API, not copy-pasted PDFs.

install · macOS & Linux
$ curl -fsSL https://github.com/guardixdev/guardix/releases/latest/download/install.sh | sh
01In the terminal

The state of your audit, without leaving the shell.

Inside a checkout of a connected repository, the CLI infers the repo from the git origin and the branch from HEAD when it is on the remote — guardix audit start needs no --repo. It asks before auditing a non-default branch or unpushed work.

guardix status example
$ guardix status
acme/vault-core
Latest audit: #4 (e7b2f4a) completed
Findings: 5 open / 23 total (1 critical, 2 high, 2 medium)
Verdict: requires action
Next: guardix finding list · guardix audit get
guardix finding get VAU-3 example
$ guardix finding get VAU-3
VAU-3 Reentrancy in Vault.withdraw() before balance update
Severity: high
Status: confirmed
Review: open
Component: Vault.sol
Description:
withdraw() sends ETH to msg.sender before writing the
updated balance, so a re-entering caller can withdraw
twice against the same recorded balance…

Output shown is illustrative, not a customer audit. Add --json to any command when it's an agent reading instead of you.

02Agent skills

Your agent becomes part of the audit.

The skills tell a coding agent when to reach for the CLI and which commands are real — so it works from live audit state instead of guessing.

finding list --json Triage findings against your diff
Your agent pulls the open findings for the repo it's sitting in and reads each one against the change it just made — from the authenticated API, not a pasted report.
finding review Dismiss with cited evidence
guardix finding review VAU-3 --status mitigated --note "…" records the verdict and the reasoning. The note lands next to the finding in the dashboard, so reviewers see why.
audit start · wait Start and watch audits
Kick off an audit on the current checkout and block until it lands. wait exits 0 on completion and 2 on timeout — a contract both scripts and agents can branch on.
cross-audit-triage Reconcile multiple audit reports
Hand your agent a Guardix run plus an external firm's report. The skill correlates duplicates and adjudicates each finding against the code on disk.
release-audit start Check a release against its audit
guardix release-audit start --report audit.pdf reviews an external audit report against the release source — the release-audit workflow, run from the terminal.
manifest Plan from the manifest
guardix manifest prints the whole CLI surface — commands, flags, env vars, exit codes — as JSON. Agents plan real invocations instead of guessing flags.
03Setup

Three steps, none of them a form.

  1. 1
    $ curl -fsSL …/install.sh | sh

    The full one-liner is at the top of this page. One static binary on your PATH; guardix doctor confirms the install.

  2. 2
    $ guardix auth login

    Prints a URL and a pairing code; you approve in the browser at guardix.io. No password ever touches the terminal. In CI, pass --api-key gdx_ak_… instead.

  3. 3
    /plugin marketplace add guardixdev/guardix
    /plugin install guardix-cli@guardix

    Run inside Claude Code. Using Cursor? The same skills work through Cursor's skills support.

guardix
read audit results and release audits
running-audits
start and watch audits
cross-audit-triage
reconcile findings across audit providers
  1. 1
    curl -fsSL …/install.sh | sh one binary on your PATH
  2. 2
    guardix auth login approve the pairing in your browser
  3. 3
    guardix doctor confirm install and auth are healthy
  4. 4
    cd your-repo any checkout of a connected repository
  5. 5
    guardix status latest audit, open findings, verdict
04Trust boundary

What it installs, and what it can never do.

A tool that sits in your shell and your agent's hands should state its own boundary. The full data-handling picture is on /security.

What the installer puts on disk
A single guardix binary on your PATH — /usr/local/bin, or ~/.local/bin when that isn't writable. The installer verifies checksums before installing anything.
How signing in works
guardix auth login prints a URL and a pairing code. You approve in the browser at guardix.io — after checking the code matches — and the CLI stores a key locally. No password ever touches the terminal. For CI, guardix auth login --api-key gdx_ak_… skips the browser.
Approval creates a named, revocable key
Approving the pairing creates an API key named "Guardix CLI". Revoke it anytime under Settings → API keys in the app, and the CLI is signed out from that moment.
The CLI never handles card data
Billing stays in the browser. guardix audit start opens Stripe checkout when payment is required, then waits quietly until you pay. --wait keeps watching the audit itself. When a release audit needs a payment or intake action, wait exits with code 3. The terminal never asks for a card. Manage invoices later with guardix billing portal.
Skills run locally
The agent skills are instructions your agent executes on your machine. They read findings over the same authenticated API the dashboard uses — scoped to the key you approved.
05Works with

Claude Code

The primary agent path — install the plugin, get all three skills.

Cursor

The same skills work through Cursor's skills support.

CI

API-key auth, --json output, and exit codes to branch on.

Any agent

guardix manifest hands any tool the full command surface as JSON.

Install it, sign in, hand it to your agent.

The whole setup is the one-liner below, a browser approval, and two plugin commands.

install · macOS & Linux
$ curl -fsSL https://github.com/guardixdev/guardix/releases/latest/download/install.sh | sh