Skill

debug

Use when encountering a bug, error, or failing test that needs structured investigation — 'this is broken', 'why does this fail', 'trace this error'. Enforces diagnosis before any fix attempt.

From dp-cto

Install

Run in your terminal

npx claudepluginhub raisedadead/dotplugins --plugin dp-cto

Tool Access

This skill uses the workspace's default tool permissions.

Supporting Assets

View in Repository

references/anti-rationalization.md

references/no-root-cause-protocol.md

references/quick-reference.md

references/red-flags.md

references/user-signals.md

Skill Content

Similar Skills

agent-harness-construction

Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.

ecc

145.8k

agent-introspection-debugging

Implements structured self-debugging workflow for AI agent failures: capture errors, diagnose patterns like loops or context overflow, apply contained recoveries, and generate introspection reports.

ecc

145.8k

agent-eval

Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.

ecc

145.8k

Stats

Parent Repo Stars3

Parent Repo Forks0

Last CommitApr 1, 2026

Actions

View Source View Plugin View on GitHub View README

dp-cto:debug — Structured Root-Cause Debugging

Read ${CLAUDE_PLUGIN_ROOT}/shared/cto-boundaries.md

The Iron Law

NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST

If you have not completed the investigation phase, you are not allowed to propose a fix. No exceptions.

Protocol

1. Gather Error Report

Collect from the user:

Error message or failing behavior (exact text, stack traces)
Steps to reproduce
Affected file paths (resolve concrete paths — never placeholders)

If the user's description is vague, ask clarifying questions before dispatch. One question per message.

2. Dispatch Debugger Agent

Dispatch debugger via the Agent tool with subagent_type: "dp-cto:debugger".

The agent prompt must include:

The error report (exact error text, reproduction steps)
Concrete file paths to investigate
The verification command to reproduce the failure

The debugger agent runs the 4-phase root cause investigation protocol (reproduce, analyze, hypothesize, report). It is read-only — it cannot modify code.

3. Validate Receipt

On agent completion, extract the structured receipt from  markers. Validate:

root_cause — file:line with explanation (not vague)
evidence — commands run and their output
suggested_fix — concrete change description
confidence — high, medium, or low

If the receipt is missing or incomplete, report the gap. Do not fabricate findings.

4. Present and Decide

Present the debugger's findings to the user with root cause, evidence, and suggested fix. Then offer next steps:

Fix it — dispatch an implementer agent (via /dp-cto:run) with the suggested fix and file scope
Investigate further — re-dispatch debugger with refined focus
Manual — user handles it themselves

Read ${CLAUDE_SKILL_DIR}/references/red-flags.md before proceeding. Read ${CLAUDE_SKILL_DIR}/references/user-signals.md before proceeding. Read ${CLAUDE_SKILL_DIR}/references/anti-rationalization.md before proceeding. Read ${CLAUDE_SKILL_DIR}/references/quick-reference.md before proceeding. Read ${CLAUDE_SKILL_DIR}/references/no-root-cause-protocol.md before proceeding.

NEVER

NEVER investigate or debug inline — always dispatch the debugger agent
NEVER propose a fix before the debugger completes root cause investigation
NEVER auto-invoke a fix — present findings and let the user decide
NEVER accept a receipt without file:line evidence
NEVER rationalize skipping process — check the anti-rationalization table
NEVER stack fixes on top of failed fixes