AxioRank for Claude Code

Zero-Trust guardrails for AI coding agents, packaged as a Claude Code plugin.

AxioRank inspects what your agent is about to do (run a shell command, write a file, query a database, post to a webhook) and what it is about to trust (tool results, fetched pages, issues) for leaked secrets, prompt injection, PII, and destructive operations. The detection engine runs locally with no API key, no network, and no signup, so you see a real result in seconds.

This repository is both an AxioRank-owned plugin marketplace and the plugin itself.

Install (about 10 seconds)

Inside Claude Code:

/plugin marketplace add AxioRank/claude-plugins
/plugin install axiorank@axiorank

That is it. Try it right away:

/axiorank:demo

What you get

Surface	What it does
/axiorank:demo	Scores five built-in attack and benign tool calls so you can watch the detectors work with zero setup.
/axiorank:scan	Inspect a tool-call JSON, a file, or pasted content on demand. Reports the decision, risk score, and every signal with redacted evidence.
Security skill	Model-invoked. Claude vets a risky action before taking it, or untrusted text before trusting it, when the context calls for it.
axiorank-security-reviewer subagent	Reviews a diff or files for secrets, PII, injection sinks, and destructive operations, confirming each finding with the engine. Run it before a commit.
Auto-scan hook	A PreToolUse hook that scans every Bash, Write, and Edit through the engine and blocks the call when it finds a leaked secret or destructive operation.
AxioRank MCP server	The hosted Zero-Trust control plane, bundled and ready. Discovery works with no key; governance tools activate when you set AXIORANK_KEY.

All of it is powered by the same published package, @axiorank/sdk, which the plugin invokes through npx. Nothing is vendored from a private repo.

The auto-scan hook

Once the plugin is enabled, the hook runs before every Bash, Write, and Edit. It builds a tool call from the content, runs it through @axiorank/sdk scan, and blocks the action only on a hard DENY: a leaked secret, a destructive operation, or risk at or above 75. Prompt injection and PII are surfaced by the skill and the subagent rather than blocked here, to keep the hook quiet on day-to-day work.

It is designed to never get in your way:

• Fail-open. Any error or timeout lets the action proceed. The hook cannot brick your session.
• No false blocks. It treats a result as a block only when the scanner actually prints a DENY, so a flaky download or an offline machine never blocks a benign action.

Knobs

Environment variable	Effect
AXIORANK_HOOK_DISABLE=1	Turn the hook off without uninstalling the plugin.
AXIORANK_HOOK_TIMEOUT_MS	Scanner timeout in milliseconds (default 10000).
AXIORANK_SDK_BIN	Override the scanner command.

Speed tip: the first npx run downloads @axiorank/sdk (a few seconds, then it is cached). For the snappiest hook, install the CLI globally and the hook calls it directly with no npx overhead:

npm install -g @axiorank/sdk

To disable just the hook permanently, you can also remove hooks/hooks.json from your installed copy, or disable the plugin with /plugin.

Optional: connect the hosted control plane

The bundled MCP server (axiorank) gives Claude AxioRank's governance tools: score a tool call against your own policy with an audit trail and approvals, verify a remote MCP server or agent card before you trust it, quarantine an agent, and more. Discovery works with no key. To enable the tool calls, create a free key at https://www.axiorank.com and export it before starting Claude Code:

export AXIORANK_KEY=axr_live_xxxxxxxxxxxxxxxx

Repository layout

.claude-plugin/marketplace.json   the marketplace catalog
plugins/axiorank/                  the plugin
  .claude-plugin/plugin.json       plugin manifest
  skills/                          /axiorank:scan, /axiorank:demo, security skill
  agents/                          the security-reviewer subagent
  hooks/hooks.json                 the PreToolUse auto-scan hook
  scripts/scan-hook.mjs            the hook implementation (fail-open)
  .mcp.json                        the bundled AxioRank MCP server

Develop and validate

Test a local copy without installing:

claude --plugin-dir ./plugins/axiorank

Validate the marketplace and the plugin:

claude plugin validate .
claude plugin validate ./plugins/axiorank

License

MIT. See LICENSE.

AxioRank: https://axiorank.com