What problem does it solve?
Getting a second opinion on code or research usually means pasting snippets into a chat, losing repository context, or risking an external agent mutating your live checkout. This Skill runs Claude Code against an isolated, editable replica of your repository so it can investigate, experiment, and report back without touching your working tree.
Core Features & Use Cases
- Sealed laboratory snapshots: Clones the current repository state (including uncommitted changes and untracked files) into an independent replica with the Git remote removed and network egress disabled.
- Checkpoint-based reporting: Claude writes structured checkpoints (
state: working, needs-decision, or complete) that you can poll without interrupting the running session.
- Follow-up conversations: Continue the same laboratory session after a decision checkpoint, archiving prior results, or start a fresh snapshot when the live checkout has moved on.
- Use Case: Ask Claude to independently review a proposed storage-layer refactor; it runs tests and commits experiments in its replica while you keep working, then returns a recommendation with cited evidence.
Quick Start
Ask the assistant to consult Claude about your design question and wait for its checkpointed recommendation before deciding what to change.
