Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Verify Before Done 1

    2

    Never mark coding work done without disk evidence plus lint, tests and build.

    Free

    1 useSecurity scanned

    verify-before-done

    Example session with this skill installed

    I've finished the user authentication refactor. I'm ready to submit the PR now.

    • Read your context and instructions
    • Compiled the verify-before-done

    I have verified the changes for the user authentication refactor against the disk and local test suite. I am now closing the task.

    Verification Report

    Change CategoryDetailsEvidence
    Modified Filessrc/auth/service.ts, src/auth/types.tsgit status confirmed
    Logic UpdateUpdated JWT validation logicsrc/auth/service.ts:42
    Type SafetyExported new AuthSession interfacesrc/auth/types.ts:12
    LintingESLint passed with 0 errorsnpm run lint output
    Unit Tests14/14 tests passedjest src/auth output
    Build StatusProduction build successfultsc --noEmit exit code 0

    Next steps

    1. Review the specific line citations in src/auth/service.ts to confirm the logic gate.
    2. Proceed with the PR merge as the build is verified green.
    3. Monitor the staging environment for integration-level side effects.

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

    About this skill

    A verification gate for AI coding agents, distilled from AgentDevs (a local-first coding harness, 12/12 in-house pilot tasks).

    The problem it solves: coding agents say "done" from memory — unverified claims, skipped test runs, tests edited to pass, whole-file rewrites that break builds.

    What it enforces: no completion claim without file:line evidence observed via read tools, the repo's real lint/tests/build run in the same turn, surgical fixes on red (never test edits, never full rewrites), and a structured closing block (changed files + citations + verify result).

    Stack-agnostic: it uses whatever lint/test/build commands the buyer's repo already has (npm, uv, gradle, maven). No MCP servers, no API keys, no network needed.

    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.

    Maximum of one automated fix attempt per cycle. Will not modify existing tests to make them pass. Requires a local verify entry point (lint/test/build).

    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

    1 use

    Times used to date

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 26 days ago

    • Free with an Agensi account

    Permissions

    Terminal / Shell
    Write Files
    Read Files

    A code repo with runnable lint/test/build commands (npm, uv, gradle, maven, …)

    Listed26 days ago

    What's inside

    Frequently Asked Questions

    It solves the problem of 'hallucinated success,' where an agent claims a task is finished based on its internal logic even if the code fails to compile or doesn't actually exist on the disk.

    The skill is framework-agnostic and works with any project that has a command-line entry point for testing, linting, or building, such as npm, pytest, or make.

    No, the skill is designed to distinguish between pre-existing errors and new regressions, ensuring the agent only focuses on fixing issues introduced by its current task.

    If a verification step fails, the agent will attempt one targeted fix; if it still fails, the agent is prevented from closing the task and must report the specific error for human intervention.

    You receive the core verification logic, integration scripts for your agent's environment, and documentation on how to hook the skill into your existing CLI tools.

    New in Testing & Debugging

    Free