AI Agent Hub
Back to skills
Debug Ultra icon

Debug Ultra

Development Updated 2026.08.29

Paste the following prompt into your AI chat to install this skill:

Please install @user_dc515faf/debug-session according to https://skillhub.cn/install/skillhub.md.

About this skill

Problem

Debugging sessions often fail because the engineer jumps to random commands, forms untested hypotheses, or chases one-off log lines for intermittent failures. Debug Ultra constrains the work into a sequence: build a trustworthy feedback loop first, generate falsifiable hypotheses, then make a minimal fix with regression checks. It explicitly blocks hypothesis work when no reliable loop exists, which prevents time spent on unconfirmed root causes.

How It Works

  • Zoom out: map relevant modules, caller chains, data shapes, state transitions, and external dependencies before diving in.
  • Diagnose: run reproduce, hypothesise, instrument, fix, and cleanup in order.
  • Feedback loop first: prefer failing tests, curl/HTTP scripts, CLI snapshot diffs, headless browser scripts, mock harnesses, fuzz loops, bisection, trace replay, or differential loops.
  • Hypothesis format: each hypothesis should read as "If X is the cause, then changing Y will make the bug disappear" and be ranked by verifiability.
  • Cleanup constraints: rerun the original repro loop, remove [DEBUG-...] logs and temporary harnesses, and record the winning hypothesis and open issues in a handoff note.

Boundaries

It fits debugging in a repository, runnable service, or scenario where request logs can be captured. If a loop cannot be built, it asks for environment access, HAR/log/core dump artifacts, or permission to add temporary instrumentation. Its command toolbox mainly covers JavaScript/TypeScript, Python, Swift, CSS, and network diagnosis, and is not a substitute for architecture review or performance tuning without measured evidence.

Use Cases

  • Build a loop, reproduce, hypothesize, instrument, and fix an API data-shape bug.
  • Reproduce intermittent Python 500s with tests or curl scripts, then rank causes.
  • Use git bisection to isolate a regression and add a regression test.
  • Write a handoff note with repro, hypothesis, fix, and open issues.

Best For

  • Full-stack engineers who need intermittent API failures converted into reproducible fixes.
  • Backend engineers maintaining Python microservices who need scripts and logs to build a loop.
  • Developers inheriting legacy code who need caller-chain and module-boundary maps first.
  • Engineers handing off incidents who need a note with repro, hypothesis, fixes, and open issues.