juliusbrussee/investigate-firstv1.0

investigate-first

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

v1.0Latest

New~161Updated Aug 17, 2026

Investigate first

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.

Files2

2 files · 1.6 KB

Select a file to preview

Overall Score

72/100

Grade

Good

Safety

Quality

Clarity

Completeness

Summary

This skill teaches agents to diagnose failures methodically before editing code. It prioritizes evidence-gathering, symptom vs. cause separation, and hypothesis ranking to avoid speculative fixes. The skill is read-only and requires no file modifications, shell execution, or external access.

Detected Capabilities

information gatheringanalysishypothesis formationevidence evaluationread-only filesystem access for diagnostics

Trigger Keywords

Phrases that MCP clients use to match this skill to user intent.

diagnose intermittent failuresinvestigate unknown errorsfind root causedebug performance regressiontroubleshoot flaky tests

Use Cases

Diagnose intermittent or flaky test failures by collecting reproduction evidence
Investigate performance regressions with profiling and metric analysis before optimization attempts
Trace root causes of unknown failures through log analysis and state inspection
Rank competing hypotheses about a bug using cheaply-falsifiable tests
Separate user-reported symptoms from implementation details to identify the actual problem

Quality Notes

Skill provides clear methodology without overspecifying tools — leaves agent flexibility in *how* to investigate while constraining the *goal*
Strong emphasis on evidence-based reasoning and avoiding speculative fixes is a positive quality signal
Instructions are concise but may benefit from example scenarios (e.g., 'if hypothesis is async race, falsify with deterministic retry loop')
No guardrails or error-handling guidance — skill assumes agent has access to relevant logs and diagnostics
Missing concrete examples of hypothesis ranking or evidence sufficiency thresholds
Interface file is minimal but correctly configured for OpenAI agents

Model: claude-haiku-4-5-20251001Analyzed: Aug 17, 2026

Reviews

Add this skill to your library to leave a review.

No reviews yet

Be the first to share your experience.

Use juliusbrussee/investigate-first in your dev environment — a Developer account adds skills to your library and syncs them via the SkillRepo CLI.

Start a Developer trial