Carveout Gate

    1

    Before your agent calls a TypeScript change done: what the change reaches, the exact tests to run, and a hold on weakened or skipped tests, CI and config edits, and new suppressions.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    carveout-gate

    Example session with this skill installed

    I fixed the discount bug. Check my change before I call it done: what does it reach, and which tests should run?

    • Read your context and instructions
    • Compiled the carveout-gate
    $ carveout impact
    2 script file(s) changed since HEAD; 1 file(s) import them, 1 test(s) among them and the change (graph 15 ms, types 178 ms).
    
    Exports whose type changed or that are gone:
      (none)
    
    Importers, nearest first:
      src/cart.ts  (1 hop)
    
    Tests that run the changed code:
      test/price.test.ts  (node, changed)
    
    Run them:
      (cd . && node --experimental-strip-types --test test/price.test.ts)
    
    $ carveout gate
    2 changed file(s) since HEAD; 1 held for review:
      test/price.test.ts: removes or rewrites 1 assertion of a test, and skips 1 more test, and rewrites or removes 1 other statement of a test (the data, setup, mocks or helpers its assertions rest on): review it and pass --allow to carry it
    exit 6
    

    The gate held it: the change skipped the test that rejects a 120% discount instead of keeping the range check. Undo the skip, restore the check, and run the one test file impact named. Only src/cart.ts imports price.ts, so nothing else needs a run.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A coding agent that cannot make a test pass has easy ways out: skip the test, loosen its assertion, mock the module under test, add a @ts-ignore, or edit the test runner's config. Its report still says all green. carveout-gate checks the change itself, from Git, before anyone calls it done.

    What it does

    • carveout impact lists the files the change reaches, nearest importers first, the exports whose type changed and the files that take them (those may no longer compile), and one command per package that runs just the affected tests. --run runs them when you agree.
    • carveout gate exits 6 and names each hold with why: deleted tests, helpers or fixtures; assertions removed, rewritten, skipped, focused with .only or swallowed by a catch; a new mock of the code's own module; code that could pass a test without the code passing it (an early process.exit, a check for a test run); edited tsconfig, linter or test-runner config; protected paths such as CI workflows; new suppressions (@ts-ignore, eslint-disable, coverage ignores); and with --scope, files outside it.
    • Runs as a Claude Code Stop hook: carveout gate --hook sends the holds back to the agent once per stop, so it fixes them before it reports.

    Neither command runs the repository's code, except impact --run. A hold the task asked for (the behaviour an assertion held was meant to change) is kept and named; --allow lets a reviewed file through.

    Requirements

    Node 22.12 or later, git and npm, in a TypeScript repository. The skill runs the open-source carveout CLI from npm (MIT) and asks before installing it. No model calls.

    Limits

    It traces imports and names, not runtime: a module loaded by a computed name, or a file read by path, is not traced. Tests under node:test, Vitest and Jest are named and run; another runner falls back to the package's npm test.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Testing & Debugging.

    Static import tracing only, no runtime path analysis. Native support limited to node:test, Vitest, and Jest. Requires Node 22.12+.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 6 days ago

    • Free with an Agensi account

    Software
    Node.jsGitnpmcarveout

    Permissions

    Terminal / Shell
    Network Access
    Read Files

    Terminal: runs the open-source carveout CLI (MIT, npm), pinned to a version; the skill asks before installing it. Read files: it reads the change from Git and the repository's import graph; neither command runs the repository's code, except impact --run when you agree. Network: the npm registry, to install the CLI. No model calls.

    Listed6 days ago

    What's inside

    Frequently Asked Questions

    A review reads the diff; carveout reads the change against the repository. impact follows the import graph and export types to the files and tests a change reaches, including ones the diff never shows. gate checks the change for ways a test can be made to pass without the code passing it: skipped or deleted tests, rewritten assertions, new mocks of the code under test, config edits and suppressions. Both are deterministic: no model calls.

    The skill is optimized for Node.js 22.12+ and offers native integration for Vitest, Jest, and the built-in node:test runner, with a generic fallback for other NPM-based testing frameworks.

    It holds, rather than judges quality: as a Claude Code Stop hook, carveout gate --hook sends the holds back to the agent (exit 2) once per stop, so the agent undoes them or names the ones the task asked for. Anywhere else, gate exits 6 with the list, so it can also run in CI or a pre-commit step.

    The tool focuses on static imports; it currently cannot trace dependencies loaded via dynamic `import()` calls with computed variables or non-standard runtime path mapping.

    It does not judge whether the code is right; it holds changes to the checks. Deleted tests, helpers or fixtures; assertions removed, rewritten, skipped, focused with .only, made unreachable or swallowed by a catch; a test's data or setup rewritten; a new mock of the code's own module; code that tells a test run from a real one; edited tsconfig, linter or test-runner config; protected paths such as CI workflows; new @ts-ignore, eslint-disable or coverage ignores; and with --scope, files outside it.

    New in Testing & Debugging

    Free