AxioRank for coding agents
Govern the agent writing your code.
AI coding agents now run commands, edit files, and call tools on your machine without waiting for review. AxioRank governs that loop. One hook for Claude Code, Cursor, Codex, and GitHub Copilot CLI blocks the dangerous actions locally and offline, reports the session to your control plane, and mints a signed, offline-verifiable proof that the code was governed.
npx -y @axiorank/coding-guard init
Choose your agent
First-class plugins for Cursor, Codex, and Claude Code. GitHub Copilot CLI is covered by the same hook via npx @axiorank/coding-guard init.
What an ungoverned coding agent can do
Destructive commands
An autonomous agent runs rm -rf, git push --force, DROP TABLE, or curl piped into a shell on your machine, faster than you can read the diff.
Secret exfiltration
A generated command or file write ships an API key, a token, or your .env to somewhere it should never go.
Agent hijacking
A poisoned MCP reply or a fetched page carries hidden instructions that steer the coding agent, the attack class researchers call agentjacking.
Watch the guard decide
This runs the real detection engine in your browser, the same one the hook runs on your machine. Pick an action, or edit the command, and see the verdict the guard would return to the agent. No API key, no network.
Edit the command. The guard re-scores it live, offline, with no API key.
A destructive operation was detected.
- Recursive/forced file delete
destructive - Dangerous CLI flag
destructive
Blocks locally, offline, with no key
The guard runs the exact AxioRank detection engine your production agents use, in the hook, on your machine. It denies destructive operations, secret exfiltration, and prompt-injected results before they run. No account required for blocking.
The same control plane as production
With a key set, every governed call flows into the same policy, audit log, alerts, and approvals as your production agents. Your org policy applies in the editor, and it can only tighten the local decision, never loosen a local block.
A signed seal nobody else mints
When the session ends, AxioRank mints a Coding Session Seal: an Ed25519-signed, offline-verifiable proof of how many calls were governed, how many were blocked, and a Merkle root over the audit trail. Provenance that the code was governed, not just a dashboard.
How it works
- 1
Install the hook
Run npx @axiorank/coding-guard init in your repo. It wires the guard into Claude Code, Cursor, Codex, and GitHub Copilot CLI.
- 2
Code as usual
Every shell command, file write, and MCP call the agent makes is scored. Dangerous ones are blocked or held for your review in the editor.
- 3
Keep the proof
At the end of the session a signed seal lands in .axiorank/. Verify it offline with npx @axiorank/audit-verify coding-seal, in CI or on any machine.
Questions
Does it block before the command runs?
Yes. The guard runs in the agent's PreToolUse hook and returns a deny verdict before the tool executes. It runs locally and offline, so blocking never depends on a network call.
Which agents are supported?
Claude Code, Cursor, OpenAI Codex CLI, and GitHub Copilot CLI. One npx command wires up all four. Codex shares Claude Code's hook schema; Copilot has its own.
Do my secrets or code leave my machine?
Not for local blocking, which needs no key and sends nothing. If you add an API key for central reporting, only the redacted payload is sent, the same as the SDK. Detector findings redact secrets in place.
What is the Coding Session Seal?
An Ed25519-signed, offline-verifiable attestation of how many tool calls a session governed, how many were blocked, and a Merkle root over the audit trail. Anyone can verify it against the published key with no trust in AxioRank.
What does it cost?
Local blocking is free on every plan, with no account. Reporting a session to your workspace is free too; it counts against your plan's normal event quota like any other governed call. The signed Coding Session Seal is the upgrade, on the Team plan and above.
Blocking is free. Proof is the upgrade.
Local guardrails for Claude Code, Cursor, Codex, and Copilot CLI are free on every plan, and so is reporting a session to your workspace (within your normal event quota). The signed Coding Session Seal is the upgrade, on the Team plan and above.