Back to skills

Agent Skill

Investigate First

investigate-first

Diagnose ambiguous failures before editing. Use for unknown causes, intermittent behavior, performance regressions, or investigations needing evidence-ranked hypotheses.

JuliusbrusseeGeneralGoAiAnthropicCavemanClaudeClaude-codeLlmMemePrompt-engineeringSkillTokens

59K installs

juliusbrussee/caveman

by Juliusbrussee

Score

8.5

/ 10

Installs

59K

Repo Stars

106.6K

Last Updated

0d ago

Fresh

Quality Ratio

86%

Description

Verified

Language

Go

First Published

Aug 2026

Summary

The Investigate First agent skill directs an AI agent to methodically diagnose ambiguous failures by gathering evidence before suggesting any code changes. This agent skill is invaluable for developers tackling complex debugging scenarios, especially when confronted with intermittent behavior, performance regressions, or failures with unknown root causes. It is a widely adopted skill in the registry, representing a skill with 12K installs. It instructs the agent to separate observed symptoms from inferred causes, trace inputs and state transitions, and rank hypotheses by evidence. Crucially, it mandates that no product code is edited until a credible mechanism explains the evidence, and exploration stops once the cause or blocker is identified.

Skill Definition

Gather evidence before changing product code.

  • Separate observed symptom from inferred cause.
  • Trace inputs, state transitions, ownership boundaries, and failure output.
  • Rank hypotheses by evidence and cheap falsification value.
  • Do not edit until one credible mechanism explains evidence.
  • Stop exploration when evidence is sufficient to name cause or exact blocker.

Report cause and proof. Make no fix unless task authorizes implementation.

How to Use

Use in O-mega

Claude Code

npx skills add juliusbrussee/caveman investigate-first
Investigate First | Agent Skills | o-mega