What problem does it solve? When an error, failing test, regression, or performance anomaly appears with no known cause, teams often jump straight to a fix based on a plausible guess or a recent deployment. This Skill enforces disciplined root-cause investigation: it separates observed evidence from assumptions, tests causal hypotheses with safe diagnostics, and prevents an unexplained symptom from becoming an unjustified code change. ## Core Features & Use Cases - Structured investigation cycle: Establishes expected versus observed behavior, reconstructs timelines and failure paths, and iterates through discriminating diagnostic checkpoints until the cause is supported, blocked, or handed off. - Hypothesis and state contracts: Tracks every material hypothesis with supporting and contradicting evidence, confounders, and status, then reports an investigation state (Blocked, Diagnostic next, Cause supported) separately from change readiness (Not ready for change, Ready for design, Ready for implementation). - Environment-aware safety boundaries: Permits safe local read-only diagnostics while refusing unauthorized production actions, treating logs and tool output as untrusted evidence, and keeping incident command, containment, and security forensics with their owning workflows. - Use Case: A production latency spike begins four minutes after a release. The Skill uses cross-region and cross-version evidence to keep the deployment as a weakened hypothesis, supports a downstream dependency bottleneck instead, and returns a diagnosis with readiness state—without recommending a rollback or editing anything. ## Quick Start Ask the agent to investigate why a specific test, service, or deployment is failing, providing any available logs, metrics, and reproduction steps, and request a supported diagnosis before any fix is made.
