debugging

mcp-debugger agent skill

This directory is a self-contained agent skill that teaches AI coding agents how to debug effectively with the mcp-debugger MCP server: when to debug instead of print, the session golden path, root-cause discipline, attach/remote recipes, and per-language quirks.

The MCP server itself exposes only tools (plus a short instructions string and a debugging-workflow prompt as in-band pointers). The full procedural knowledge — workflow, bisection discipline, language references — lives here, where every agent harness can load it.

Install

Claude Code (project- or user-level):

# user-level: available in every project
mkdir -p ~/.claude/skills && cp -r skills/debugging ~/.claude/skills/mcp-debugger

# project-level: committed with your repo
mkdir -p .claude/skills && cp -r skills/debugging .claude/skills/mcp-debugger

Cross-agent directories (GitHub Copilot CLI, and harnesses that read ~/.agents/skills/):

mkdir -p ~/.agents/skills && cp -r skills/debugging ~/.agents/skills/mcp-debugger
mkdir -p ~/.copilot/skills && cp -r skills/debugging ~/.copilot/skills/mcp-debugger

Any other agent: paste the contents of SKILL.md into your system prompt or rules file (e.g. .cursorrules, Cline custom instructions), and keep the references/ files reachable so the agent can read the per-language guides on demand.

Contents

• SKILL.md — entry point: when to debug, golden path, root-cause discipline, attach recipes, current limitations
• references/python.md, javascript.md, ruby.md, rust.md, go.md, java.md, dotnet.md, cpp.md — per-language prerequisites, quickstarts, quirks, troubleshooting

Keeping it honest

The skill states current limitations plainly (e.g. adapter-specific pause_execution support, Ruby attach stdout). If you hit a behavior the skill doesn't describe, please open an issue — the skill is maintained against real agent transcripts.