Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Evidence-Gated Completion

    1

    Stop false 'done' reports: gate every completion claim behind fresh test runs, exit codes, and diffs, so nothing gets merged or closed on hope alone.

    Free

    1 useSecurity scanned
    Evidence-Gated Completion

    Evidence-Gated Completion

    Example session with this skill installed

    I just finished fixing the token refresh bug in the auth-service repo and my agent already says 'Great, fixed and ready to merge!' I want that claim audited before I approve the PR. Please walk every statement in its completion report through the verification gate: re-run npm test, lint, typecheck, and build from the current tree, confirm the reproduction case for issue-412 actually passes, review git diff against origin/main, and tell me which claims are verified, which are only partially covered, and which must be removed from the report until real evidence exists.

    • Read your context and instructions
    • Compiled the evidence-gated completion

    Summary

    Verification gate applied to the checkout-fix completion report before merge. Three of four success claims were confirmed with evidence produced after the last edit; the fourth was downgraded to unverified and stripped from the 'ready' language.

    ClaimProof run after last editOutput readStatus
    Bug #412 fixednpm test -- repro/issue-4121/1 passed, exit 0Verified
    All tests passnpm test87/87, 0 failuresVerified
    Build succeedsnpm run buildexit 0, 0 warningsVerified
    Checkout works end-to-endnot run (no staging creds)noneUnverified

    Original report said 'fixed and ready to merge'; corrected status: verified for unit and build scope only, the end-to-end checkout flow remains unproven.

    Next steps

    • Run the Playwright checkout flow on staging with real credentials before marking the PR ready
    • Lock the fix in with a red/green regression pass: revert the patch, confirm issue-412 fails, restore, confirm it passes
    • Record the unverified end-to-end item in the PR description as a known gap

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

    About this skill

    Stop reporting work as "done" based on assumptions or stale data. This skill enforces a strict evidence-based protocol for developers and agents to verify code changes, bug fixes, and deployments. It replaces vague claims like "it should work" with a mandatory verification gate that requires fresh command output, successful exit codes, and diff reviews after every change.

    What it does

    • Enforces verification gates by requiring fresh proof before any success claim is reported to the team.
    • Identifies language traps by flagging speculative words like "probably" or "seems" that signal missing evidence.
    • Maps claims to proof by defining specific requirements for test passes, linting, build success, and UI changes.
    • Automates regression testing using a red-green-pass pattern to ensure a fix actually addresses the root cause.
    • Evaluates subagent output through mandatory diff reviews and independent command execution rather than blind trust.

    How it works

    1. State the exact claim being made, such as "the bug is fixed" or "the build passes."
    2. Run the specific command, test, or query required to prove that claim on the latest code version.
    3. Inspect the full output, including exit codes and failure counts, to ensure the result matches the claim.
    4. Report the status as verified, partially verified, or unverified based on the captured evidence.

    Frameworks & tools

    This skill is platform-agnostic and works with any CLI-based toolchain including npm, pytest, make, git, eslint, and docker. It integrates into standard CI/CD workflows and local development environments.

    Why this beats prompting it yourself

    Generic prompts often allow agents to hallucinate success or rely on logs from five minutes ago. This skill embeds a rigid "evidence-first" logic that prevents the agent from moving to the next task until it provides proof from the current state.

    Use cases

    • Pull request preparation to ensure every commit passes linting, typing, and testing locally before pushing.
    • Bug fix validation to confirm a regression test fails without the fix and passes with it.
    • Subagent oversight to audit changes made by other AI tools through direct diff inspection and verification.
    • Deployment safety to catch environmental issues where a build succeeds but runtime execution is unverified.

    Known limitations

    Verification requires a functional environment where tests or builds can execute. If external dependencies or credentials are missing, the skill defaults to "unverified" status.

    Requires the project to have runnable checks; claims about hardware, production-only systems, or credential-gated services can only be reported as unverified. It also adds a small overhead to trivial edits where a full gate is overkill.

    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 any payment.

    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
    Browser

    Only shell access is needed to re-run the project's own test, lint, build, and git commands; nothing leaves the machine.

    Listed26 days ago

    What's inside

    Frequently Asked Questions

    This skill acts as a strict verification layer that prevents agents from reporting tasks as 'complete' based on assumptions; it requires real-time proof like exit codes and diffs before moving to the next task.

    It is platform-agnostic and works with any CLI-based toolchain (e.g., npm, pytest, git, docker), making it compatible with any agent capable of executing terminal commands.

    You receive the core logic framework for the evidence-gated protocol, integration instructions for your agent's workflow, and templates for defining verification requirements across different languages.

    No, this skill is designed to run in real-time within your active development environment to ensure it captures fresh command output and current file states.

    If the skill cannot execute the required tests due to missing dependencies, it will default to an 'unverified' status rather than making a guess, ensuring no false positives are reported.

    No. It re-runs whatever commands the project already uses (npm, pnpm, pytest, make, cargo, etc.) and judges the exit code and failure counts of that run.

    The gate targets only the claims being made, so a docs edit needs a docs check, not a full CI pass. Large or merge-level claims get the full sequence.

    Yes. It includes a delegation pattern that reads the actual git diff and independently re-runs the relevant checks instead of trusting the agent's own 'done' summary.

    New in Testing & Debugging

    Free