Bug investigation & fix

Assign a bug and a coding agent investigates the root cause, proves it with a failing test, fixes the smallest surface, self-reviews the diff, and drafts a pull request for you to merge.

Join the waitlist

What it installs

Agents 1

  • Coding Agent

    Fixes assigned bugs in the connected repository with a proven regression test and a self-reviewed PR.

Workflows 1

  • Diagnose, fix, self-review, draft PR

    Build a feedback loop, find the root cause, fix it with a test, self-review the diff, and draft a PR for approval.

Goals 1

  • Bugs closed with a proven, self-reviewed fix

    Reported bugs are resolved at the root cause, locked down by a regression test, and shipped through a self-reviewed PR.

Skills 4

  • diagnosing-bugs

    Feedback-loop-first diagnosis for hard bugs and performance regressions: build a tight, deterministic, red-capable signal before forming any theory. Adapted from mattpocock/skills/diagnosing-bugs.

  • systematic-debugging

    A disciplined four-phase debugging loop governed by an iron law — no fix without root-cause investigation first — escalating to architecture after three failed fixes. Adapted from obra/superpowers/systematic-debugging.

  • find-bugs

    Branch-diff review for bugs, security vulnerabilities, and quality issues: gather the full diff, map the attack surface, run a per-file security checklist, verify each finding is real, and audit before concluding. Adapted from getsentry/skills/find-bugs.

  • django-perf-review

    Validated performance review (N+1 queries, unbounded querysets, missing indexes, write loops) where every finding is traced and proven before reporting, and severity must match real impact. Adapted from getsentry/skills/django-perf-review.

Requirements

What this template expects to do its job. Task Machine does not verify these — you decide whether your setup is ready.

  • Connected repository — Needs a connected code repository to read the failing code, build a feedback loop, and propose a fix.
  • Browser access to GitHub — The agent works in GitHub through its web interface in your browser and pauses for your approval before making any changes. Until you connect it, it works from attached exports.
  • A runnable test or repro harness — The loop needs a way to run the bug code path unattended — a test runner, a dev server to curl, or a CLI invocation — so the fix can be proven, not assumed.

Get started

Install Bug investigation & fix and run it with approvals.

Join the waitlist and we will send early access when the first private beta spots open.

Private beta. We invite teams in batches and never share your email.