Scopebond — AI skill for Claude Code

Scopebond

[](LICENSE) [](https://github.com/avouro-com/scopebond/actions/workflows/ci.yml) [](https://www.npmjs.com/package/@scopebond/gateway) [](https://www.npmjs.com/package/@scopebond/hook)

**Scopebond is an open-source checkpoint for AI coding agents (Claude Code, Cursor, Codex, MCP, GitHub) that blocks out-of-policy actions before they run and signs every decision into a tamper-evident record anyone can verify offline.** It checks every action a coding agent takes — a shell command, a file edit, a git push, an MCP tool call — against rules you write, blocks the ones that break them, and records the rest.

Try it: [scopebond.com](https://scopebond.com) · live policy demo at [try.scopebond.com](https://try.scopebond.com).

**Experimental alpha — use controlled test systems only.** The semantics below are the target; the alpha is under active hardening. The hosted workspace ("Scopebond Cloud") is a separate, proprietary product and is not in this repository.

Who it's for

• A developer running Claude Code, Cursor, or Codex who wants a guardrail that blocks a destructive command before it runs, not a log they read afterward.
• A team lead who has to show what the agents did — which actions, decided by which rule — to a client, an insurer or an auditor.
• An auditor or client who needs to check that record independently, offline, with no account and no trust in the vendor.

60-second start: govern your coding agent

Put Scopebond in front of Claude Code, Cursor, or Codex. One command scaffolds a signing key and a starter policy and wires the hook into your agent:

npx @scopebond/hook init            # Claude Code; add --cursor o