More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    The Tracker: prove the root cause before any fix

    1

    Your agent stops blaming the last commit. It proves the cause with a failing repro and a disproved theory before it touches the code.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    The Tracker: prove the root cause before any fix

    The Tracker: prove the root cause before any fix

    Example session with this skill installed

    Customers report wrong invoice due dates since last week's deploy. An invoice issued 2026-03-31 with 1-month terms shows due 2026-04-28, it should be 2026-04-30. I am pretty sure the last commit (the Decimal rounding change) broke it, so maybe just revert it. Find the real root cause and fix it.

    • Read your context and instructions
    • Compiled the the tracker

    theory: the Decimal rounding commit (HEAD) broke due dates
    check FAIL (predicted pass) on HEAD~1 (59d2915b): DISPROVED
    first commit where the repro fails: 3d7ce6fa perf: drop calendar import from add_months
    files: billing/dates.py

    • Before the fix: failed (exit 1) on 186eb073
    • After the fix: passes on HEAD 4932c2a9
      3 theory check(s), 1 disproved.

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

    About this skill

    Follows tracks, not rumours. The last commit is innocent until proven guilty.

    Coding agents debug by guessing. They blame the most recent commit, patch the line where the error shows up, see one green run and say "fixed". The real cause is often older, and it only shows under one condition.

    The Tracker fixes the order of work. Before any code changes, the agent:

    1. Writes the bug as a command that fails today. The script refuses a repro that passes.
    2. Writes down the trigger, the condition that makes it happen only sometimes, and the symptom, as three separate facts.
    3. Runs the repro on a working path next to the failing one: another input, another env var, or another commit in a temporary git worktree. An optional git bisect run names the first failing commit.
    4. States each theory so a command can disprove it, runs that command, and records "survived" or "disproved".
    5. Shows the diagnosis to you and waits for your yes. Then it fixes the cause, and the repro becomes the regression test.

    What you get: a .rootcause/<slug>.json record and a Markdown block for the PR. It shows the repro failed before the fix and passes on HEAD, the trigger, condition and symptom, every comparison, the bisect result, and which theories were disproved. Worktrees and bisect state are removed after every run, also on errors and Ctrl+C.

    In our test, the user blamed the last commit. The agent showed the bug already failed on the commit before it, bisected to an older "perf" change, and fixed that instead.

    For developers who use Claude Code or other coding agents on any git repo and any language, and want bug fixes they can trust in review. One Python script, standard library only, nothing leaves your machine.


    Part of The Guild, agent skills by Sesisnando Neto: The Oathkeeper (sealed acceptance checks), The Quartermaster (quests and decisions), The Archivist (memory hygiene), The Lorekeeper (repo principles from history), The Inspector (QA validation).

    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.

    Needs a bug you can turn into a command that fails. Bugs seen only on a live system with no way to copy the data or state will need manual work first. Bisect is slow when each step runs a long build or test. Commits that do not build are skipped only when the repro times out; otherwise they count as bad, so narrow the range first. Theory checks show support or disproof, not proof. The approval step depends on the agent following the skill.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed today

    Software
    GitPython 3.8+BashvitestNode.jsjest

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    File Scopes

    .rootcause/**
    at
    the
    repo
    root
    temporary
    git
    worktrees
    in
    system
    temp
    folder

    Runs git and the project's own repro and check commands. Temporary git worktrees go to the system temp folder and are removed after each run. No network access.

    Listedtoday

    What's inside

    Frequently Asked Questions

    No. The script only runs git and the commands you or the agent give it, on your machine. It makes no network calls.

    Yes, if the bug can be written as a shell command that exits non-zero. The references include patterns for Python, JavaScript, Go, Java and Ruby.

    Runs on other commits happen in a temporary git worktree in the system temp folder. It is removed after each run, also on errors and Ctrl+C, and git bisect reset runs inside it. If the machine crashes mid-run, the cleanup command removes what is left.

    No tool can promise that. It makes the agent show its evidence: a failing repro, a working path next to it, and theories that had a real chance to be proven wrong. You still judge the result.

    The skill tells the agent to show you the diagnosis and wait for an explicit yes, and the report records who approved. You can also pre-approve in your request.

    Pass it with --with and it is copied into each temporary worktree. Folders such as node_modules are linked, not copied.

    New in Testing & Debugging