# Review Software Reliability

> Software reliability review for failure handling, timeouts, retries, idempotency, concurrency, resource safety, transactions, consistency, degradation, and recovery. Use for resilience or fault-tolerance questions.

- Skill: `kenlck/review-software-reliability` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add kenlck/review-software-reliability`
- Raw SKILL.md: https://api.skillmd.com/api/skills/kenlck/review-software-reliability/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: kenlck (https://skillmd.com/u/kenlck)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/kenlck/review-software-reliability

---


# Review software reliability

Assess one repository, a repository group, or an existing fleet evidence bundle. Trace how expected failures are contained, surfaced, and recovered.

## Run the lens

1. Read the [standalone lens protocol](../audit-software-fleet/references/standalone-lens.md) completely and prepare the supplied scope with lens ID `software-reliability`.
2. Read [the reliability checks](references/checks.md) completely. Apply every check family relevant to state changes, resources, concurrency, dependencies, and interaction edges.
3. Trace concrete failure paths and recovery behavior. Classify supported outcomes as strengths or findings; record missing runtime, transaction, or dependency semantics as unknowns and incomplete coverage.
4. Follow the protocol to account for every contract-registered check family, validate the result, ingest it, and hand back the canonical artifact path.

For raw collection or authorized failure testing, read and follow the [safety model](../audit-software-fleet/references/safety-model.md). Keep static inferences distinct from observed failure behavior.

## Boundary

Own correctness and recovery under failure. Route production visibility and response controls to operability, measured latency or resource cost to performance, and test strategy to test quality.

## Completion

**Complete when:** every in-scope repository has an exact commit or explicit unavailable state, every applicable check family is accounted for, every claim has bounded sanitized evidence at an appropriate confidence, gaps remain visible in unknowns and coverage, and the shared result validator succeeds.