What problem does it solve? Codebases encode implicit theories of failure that are hard to audit manually: swallowed exceptions, missing timeouts, non-atomic writes, and unbounded retries hide in thousands of call sites. This Skill reads the error-handling code of a project and produces a verifiable reliability report describing how the system behaves when dependencies, the network, the disk, or its own logic fail. ## Core Features & Use Cases - Error model and handling analysis: Maps exception/Result conventions, central error types, swallowed or catch-all handlers, panics on runtime input, and success-reported-after-failure patterns. - Isolation, recovery, and degradation review: Traces process/task/plugin boundaries, retries, timeouts, circuit breakers, fallbacks, and offline modes, citing the exact constants and code. - Crash safety and shutdown audit: Checks atomic writes, fsync, cleanup on failure paths, signal handling, and shutdown order, then synthesizes a reliability posture with blast-radius assessment. - Use Case: Point it at a repository with a Sokrates _sokrates analysis to receive a validated reliability-scan.json findings report plus an interactive HTML explorer, with severity-calibrated findings and stable IDs that diff across runs. ## Quick Start Ask the AI to run the reliability scan on this repository and report how the system behaves when its main dependencies fail.
