More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    duediligence

    1

    Your agent says all tests pass. Check with tests it never saw: a public suite it works against and a calibrated hidden suite that judges the result, written from a spec you confirm.

    Free

    Security scanned

    duediligence

    Example session with this skill installed

    Write tests my agent can't see for this ticket: "Accept 1h30m-style timeouts: convert human-readable durations like 1h30m, 90s or 250ms to milliseconds, and print them back in the same style."

    • Read your context and instructions
    • Compiled the duediligence

    Drafted specs/duration-parse/spec.md from the ticket, with the 9 decisions it leaves open for you to confirm, each beside the ticket sentence it answers: whitespace, upper-case units, leading zeros, unit order, a bare number, weeks and fractions, totals above Number.MAX_SAFE_INTEGER, error messages, and formatting zero.

    After you confirmed them, prepared the suites in blind sessions: 22 public tests and 39 hidden, all 39 kept by calibration against the reference, for $1.22 in about 5 minutes.

    The handout folder (public suite + BRIEF.md) is ready for your implementing agent. When it is done, run duediligence judge duration-parse <folder>/src for the verdict: passed, overfit, incomplete, regressed or refused.

    BRIEF.md

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

    About this skill

    Coding agents write the code and the tests that pass it, so a green run mostly tests the agent's own guesses. duediligence turns a ticket or spec into two test suites: a public one the implementing agent works against, and a calibrated hidden one it never sees, which judges the result.

    In the published experiment, 75 of 105 agent runs whose own tests passed failed tests they never saw. Given the public suite, the same agent passed 81 of 105 instead of 30, and a blind referee on another model upheld 95 of 100 rejections.

    What it does

    • Confirms the spec first. Each decision the ticket leaves open is put to you, beside the ticket sentence it answers, before any test is written.
    • Writes both suites in blind sessions, apart from the implementer, and calibrates the hidden suite against an independently written reference: a test the reference fails is dropped.
    • Hands out a folder with the public suite and a brief, for any implementer: Claude Code, Codex, Cursor, Copilot, Aider or a person.
    • Judges the submission: passed, overfit, incomplete, regressed or refused, with the failing scenarios.
    • Runs in CI: seal a spec's suites into one file and judge pull requests from GitHub Actions.

    Targets

    • A new module in TypeScript, Python, Go or Rust, or a command or HTTP service checked at its boundary.
    • A change to one package of an existing TypeScript or Python repository, whose existing suite must keep passing.

    Requirements

    Node 20.19 or later, and one account for the test-writing sessions: Claude Code signed in, a Codex login, or an Anthropic API key. The skill runs the open-source duediligence CLI from npm (MIT) and asks before installing it. Preparing one spec takes about $1–4 of model usage and 10–25 minutes.

    Limits

    One module or one change, about an hour of work, with no network or outside service. Rendered UIs and migrations do not fit.

    No UI/Playwright support. Scope limited to 1-hour tasks. No external network/services in tests. Windows requires WSL2 for sandboxing.

    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

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 5 days ago

    • Free with an Agensi account

    Requires

    Node.js
    Claude Code, Codex login or Anthropic API key

    Permissions

    Read Files
    Write Files
    Network Access
    Terminal / Shell
    Environment Variables

    Terminal: runs the open-source duediligence CLI (MIT, npm), pinned to a version; the skill asks before installing it. Network: the npm registry for the CLI and its test toolchain, and model sessions through the account you choose (Claude Code, Codex or the Anthropic API). Environment: ANTHROPIC_API_KEY when that is the account, DUEDILIGENCE_* settings, and optional JIRA_* to fetch a ticket. Files: specs, suites and runs under ~/.duediligence/, and the handout folder you choose.

    Listed5 days ago

    What's inside

    Frequently Asked Questions

    `duediligence` prevents agents from 'gaming' the system by creating two separate test suites; the agent uses a public suite to guide development, but its final score is determined by a hidden, calibrated suite it has never seen.

    Any coding agent that loads SKILL.md skills: Claude Code, Codex, Cursor, Copilot and others. The tests are written in sessions on the account you have (Claude Code signed in, a Codex login or an Anthropic API key); the implementer can be any agent or a person.

    The skill is one SKILL.md: the steps your agent follows. It runs the open-source duediligence CLI from npm (MIT), pinned to a version, and asks before installing it. The CLI installs its test toolchain under ~/.duediligence/ on first use.

    It supports logic-heavy modules, CLI tools, and HTTP services in TypeScript, Python, Go, and Rust, though it currently does not support frontend UI testing or browser-based automation like Playwright.

    Yes, for one package of a TypeScript or Python repository: `duediligence fit` checks whether the package can be a target, `repo` points a spec at it, and the change is built in your own checkout, where the existing suite must keep passing.

    New in Testing & Debugging