What problem does it solve? After finishing a nontrivial code change, the author is often blind to defects in their own reasoning. This Skill obtains a genuinely independent second opinion from a fresh subagent, then forces every reported finding to be verified against the real code before any fix is made. ## Core Features & Use Cases - Independent subagent review: Spawns a fresh gpt-5.6-sol subagent with no inherited context to review the diff read-only, reporting only actionable bugs, architectural problems, brittle tests, and missing coverage. - Finding verification workflow: Each finding is treated as a claim to be proven or rejected against the actual code, so fixes are made only for confirmed defects. - Validation sequencing: Runs cheap focused checks before review and defers expensive aggregate validation until after findings are resolved, avoiding wasted validation runs. - Use Case: After implementing a refactor that moves state handling between modules, run this Skill to get an unbiased review of the diff, confirm or reject each reported issue, fix the real ones, and report the disposition of every finding. ## Quick Start Ask the assistant to run an independent review of the change you just finished before declaring it done.
