Debug Deep

Tree-of-thoughts reasoning for elusive bugs.

When to Use

• Intermittent or hard-to-reproduce bugs

• Standard debugging failed

• Race conditions suspected

• Multiple potential causes

Artifacts

• DEBUG-PLAN-[BUG-ID].md

• task.md updated via task_boundary

Approach

Phase 1: False Premise Challenge

List every assumption in the bug report. Test each as True/False before proceeding:

• Assumption 1: [e.g., "Bug occurs on every load"]

• Assumption 2: [e.g., "Data is correctly formatted"]

Phase 2: Tree of Thoughts (Multi-Hypothesis)

Generate 4-6 hypotheses with dual investigation paths:

Hypothesis Confidence Path A Path B Selected

1 ... High Check X with grep Instrument Y A (faster)

2 ... Medium Trace via debugger Reproduce isolated B (reliable)

Phase 3: Investigation & Fix Plan

• Primary Fix Plan: Main approach

• Reflexion (Security Engineer): List 3 ways fix might fail, introduce vulnerabilities, or degrade performance

• Fallback Plan: If primary fails

Phase 4: Systemic Analysis

• Does this bug represent a class of problems?

• Should a new rule be added to prevent recurrence?

Constraints

• Non-Linear Exploration: Don't converge prematurely

• Must have fallback plan

• If fix fails, iterate (max 3 attempts)