CLI and agent skills
The Guardix CLI (guardix) connects your terminal — or a coding agent — to the same audits and findings you see in the web app. Guardix also ships agent skills that teach tools like Claude Code how to read results, start audits, and compare reports from multiple providers.
Install the CLI
Section titled “Install the CLI”macOS and Linux (recommended)
Section titled “macOS and Linux (recommended)”curl -fsSL https://github.com/guardixdev/guardix/releases/latest/download/install.sh | shThe installer picks the right binary, verifies checksums, and places guardix on your PATH (typically /usr/local/bin or ~/.local/bin).
Pin a version or custom install directory:
curl -fsSL https://github.com/guardixdev/guardix/releases/latest/download/install.sh | GUARDIX_VERSION=v0.3.4 shcurl -fsSL https://github.com/guardixdev/guardix/releases/latest/download/install.sh | GUARDIX_INSTALL_DIR="$HOME/bin" shManual download
Section titled “Manual download”Download a release archive from github.com/guardixdev/guardix/releases, extract, and add the guardix binary to your PATH.
Verify
Section titled “Verify”guardix --versionguardix doctorAuthenticate
Section titled “Authenticate”Browser pairing (interactive)
Section titled “Browser pairing (interactive)”guardix auth loginThe CLI prints a URL (for example https://guardix.io/cli/auth?s=…) and opens your browser. Sign in with GitHub or Google if needed, approve the pairing, and the CLI stores a key locally. The approval page also has a Deny button if the request isn’t yours. Revoke CLI keys anytime under User menu → API keys in the app.
Sign in from an agent
Section titled “Sign in from an agent”Coding agents and remote shells can’t open a browser. From CLI 0.3.4 onward:
guardix auth login --no-browser --json--no-browser prints the verification URL instead of opening a browser, and --json emits machine-readable status lines:
{"status":"awaiting_approval","verification_url":"https://guardix.io/cli/auth?s=…","user_code":"…"}{"status":"authenticated"}The agent relays the verification_url to you; open it in any browser, confirm the code matches, and Approve (or Deny) the pairing. When approved, the CLI emits {"status":"authenticated"} and stores the key. --json also implies --non-interactive, so the command never prompts.
API key (automation / CI)
Section titled “API key (automation / CI)”guardix auth login --api-key gdx_ak_...Or set GUARDIX_API_KEY in the environment. See API keys for creating keys in the dashboard.
Check status:
guardix auth statusEveryday commands
Section titled “Everyday commands”Inside a connected git checkout, many commands infer the repository from git remote:
guardix status # latest audit summary + open findingsguardix finding list # findings for the latest auditguardix finding get VAU-3 # one finding by short codeguardix audit start --ref main # start a full auditguardix audit wait --audit '#2' # block until completion (CI-friendly)Add --json for machine-readable output (it also implies --non-interactive, so commands never prompt). Run guardix manifest --quiet for the full command surface — do not guess flag names.
Useful audit flags mirror the web wizard: --contract (repeatable scope), --doc-url, --doc-text, --doc-file, and --include-repo-docs.
Billing
Section titled “Billing”The CLI never handles card data. Paying for a full audit is:
guardix audit start # opens Stripe checkout if needed and waits until you payguardix audit start --wait # then keep watching until the audit finishesguardix billing portal --open # invoices and saved payment methods after the first paymentguardix quota # free-audit remaining + per-audit pricebilling portal is available after the first payment (Stripe returns 409 until then). Release Audits that need payment or intake action exit wait with code 3 and print the checkout URL instead of holding.
Agent skills (Claude Code plugin)
Section titled “Agent skills (Claude Code plugin)”The Guardix CLI plugin bundles three skills for coding agents:
| Skill | Use when |
|---|---|
guardix | Reading audit results — status, findings, finding codes like VAU-3, review state. |
running-audits | Starting and monitoring audits (audit start, wait, get, logs). |
cross-audit-triage | Comparing findings from multiple sources (Guardix + external firm reports or tool output), correlating duplicates, and adjudicating true vs false positives locally. |
Install the plugin
Section titled “Install the plugin”From inside Claude Code:
/plugin marketplace add guardixdev/guardix/plugin install guardix-cli@guardixOr from a regular shell (scripts, CI):
claude plugin marketplace add guardixdev/guardixclaude plugin install guardix-cli@guardixSkills live in the plugin package; each skill file describes when the agent should invoke it and which CLI commands to run. Agents should prefer guardix manifest over inventing flags. The same three skills also work in Cursor, which auto-loads agent skills from its skills directories (.cursor/skills/, ~/.cursor/skills/) as well as compatibility paths like ~/.claude/skills.
Cross-audit triage
Section titled “Cross-audit triage”The cross-audit-triage skill expects report files plus a checkout of the audited commit on disk so findings can be validated against real source. It produces a comparison matrix and optional cross-audit-triage.json artifact. You can sync adjudications back with guardix finding review when ready.
Team context
Section titled “Team context”guardix team listguardix team switch <team-slug>The active team filters which repositories and audits are visible — match the workspace you use in the browser.
Related
Section titled “Related”- API keys — Bearer tokens for HTTP and CI
- Authentication — web sign-in
- Your first audit — browser-first walkthrough