Implementation Review (review-spec)

What it does

Systematically compares a feature's design and implementation documentation (design.md, implementation.md, tasks.md) against the actual implementation to detect missing implementations, extra or deviating code, and documentation inconsistencies. Produces classified findings with severity and confidence scores and supports indexing intentional deviations as architecture decisions.

When to use it

Use after implementing a feature or mid-implementation to verify conformance to design docs, before merging large changes, or when audit-level assurance is needed. Supports an isolated review mode that runs an independent sub-agent for higher rigor.

What's included

• Scripts: none embedded in this skill bundle (has_scripts=false)
• References: supporting profile and workflow docs (shared-profile-detection.md, review-process.md) are included in the sibling files.
• Instructions: step-by-step workflow for loading spec documents, detecting profiles, running per-task verification, and producing a review report with mandatory confidence reasoning. The skill specifies finding types, severity (P0–P3), and confidence scoring guidance.

Compatible agents

Best suited for Claude-Code / agentic Claude workflows, and projects using Codex-style review automation. The guidance is platform-agnostic and can be applied by other agent runtimes that can read repo docs and run sub-agents.