Debug Mode

An Agent Skill that brings Cursor's Debug Mode to any skills-compatible agent. Hypothesis-driven debugging with runtime log collection.

How It Works

       You                    Agent                    Your App
        |                       |                         |
        |   "login is broken"   |                         |
        |---------------------->|                         |
        |                       |   generates hypotheses  |
        |                       |   instruments code      |
        |                       |------------------------>|
        |                       |                         |
        |   reproduce the bug   |                         |
        |---------------------------------------------->  |
        |                       |   <-- logs flow back    |
        |                       |                         |
        |   "fixed!"            |   analyzes, patches     |
        |<----------------------|                         |

The Loop:

1. Describe the bug
2. Agent generates hypotheses & instruments code with logging
3. You reproduce the bug
4. Agent analyzes runtime logs—no assumptions, only evidence
5. Agent applies minimal fix backed by log data
6. You verify
7. Agent cleans up all instrumentation

Core principle: Never assume. Collect evidence. Validate hypotheses. Conclude from data.

Architecture

Uses tmux for server lifecycle management:

• debug-log-server - Zero-dependency Node.js HTTP server collecting logs from your app
• debug-dev-server - Your dev server (npm run dev, python manage.py runserver, etc.)

Logs POST to localhost:1337 from browser or server code. Works with JS/TS, Python, Go.

Requirements

• tmux - server lifecycle management
• node - runs the log collection server (zero dependencies)

Installation

Via npx:

npx skills add ethanfann/debug-mode

Via simba:

simba add ethanfann/debug-mode

Manual:

cp -r skills/debug-mode ~/.claude/skills/

Usage

Just describe your bug. The skill activates when debugging needs runtime inspection:

"The submit button doesn't save the form data"

"API returns 500 but only on the second request"

"State updates but UI doesn't re-render"

Why This Exists

I really like Cursor's debug mode and wanted to use it in CLI agents.

---

Part of the Agent Skills ecosystem