Plugins for Cursor, ChatGPT and Codex

One portable plugin. Your agent audits your code and adds the boundary, entirely locally.

What is inside

Part	Name	What it does
Skill	`fail-closed-governance`	Adds a deterministic boundary to an existing project
Skill	`governance-review`	Audits for fail-open paths and bypasses
Skill	`secure-agent-tool`	Wraps one function, tool or MCP operation
Rule	`fail-closed.mdc`	Keeps agents from granting themselves authority
Command	`/governance-review`	Traces proposal → authorization → effect for every path
Command	`/protect-tool`	Wraps the selected function and adds tests

What a review reports

For each consequential side effect it lists every path that can reach it and classifies each as one of:

FAIL_OPEN: an error, timeout or missing config grants permission
BYPASS: a path reaches the effect without the check
SELF_AUTH: the agent can grant itself authority or approval
STALE_AUTH: authority is cached, unexpired-forever, or not re-checked
UNBOUND_EFFECT: approval applies to free text or something that can change afterwards
MISSING_RECEIPT: the action can happen with no record
POLICY_ERROR: malformed, ambiguous or silently-defaulting policy
APPROVAL_BYPASS: high-risk action proceeds without independent approval
OK

It ends with the residual risk that remains even after fixes.

What the agent can do for you

Review a repository, identify consequential side effects and fail-open paths, propose a patch, install Core, wrap functions and tools, generate policy configuration, generate and run tests, and explain what is still risky. No account, no network calls, and no commercial service required.

Install

The plugin lives in the plugin/ folder of the repository, with a Cursor manifest, a Codex compatibility manifest and a portable Agent Plugin manifest. Marketplace listings are being prepared; until then, copy the folder into your agent's plugins location.