Guard

Recursive Guard

Keep enforcement separate from advisory Recursive capabilities. Installing this plugin does not edit a repository and does not enforce anything until the repository owner explicitly adds .recursive-guard.json and trusts the bundled hook.

Read [the policy contract](references/policy.md) before proposing integration.

Inspect first

• Preserve all existing AGENTS.md, CLAUDE.md, .codex/, .claude/, agents, skills,

and hooks.

• Check whether .recursive-guard.json already exists without reading unrelated config.
• Explain that Codex requires a separate review/trust decision for the exact hook hash.
• Never create or change the policy without explicit approval for the exact patch.

Propose the smallest policy

Use schema version 1, choose audit before enforce when the repository owner wants a trial, and list only repository-relative protected paths. The policy file protects itself.

{
  "schema_version": 1,
  "mode": "audit",
  "protected_paths": ["AGENTS.md", "CLAUDE.md", ".codex", ".claude"]
}

Present that as a reviewed patch. Do not apply it merely because the plugin is installed.

Verify

After explicit integration, exercise one allowed path and one protected path. audit emits a warning but permits the call; enforce denies supported matching tools. Confirm that existing provider configuration remains otherwise unchanged.

This is a guardrail, not a sandbox. Codex documents that some specialized tool paths may bypass lifecycle hooks. Use repository permissions, branch protection, and CI for durable enforcement.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: GhostlyGawd
• Source: GhostlyGawd/recursive-harness
• License: MIT

Install and usage instructions live in the source repository linked above.