pull-spiderman
Pull Spiderman
Drive a PR's review comments to a vetted decision and a reply. A reviewer makes a claim; we adversarially challenge it twice (once in a subagent, once on the main agent with full conversation context) before deciding whether it is right — two Spider-Men pointing until one claim is confirmed. Every reply, resolve, and code fix is gated behind explicit per-comment user approval. The reviewer may be a bot or a human; the skill treats them the same.
A PR that is one layer of a stack gets one more pass before any verdict: the layers above it may already do what the reviewer asks, and a layer that has been published is never edited (ADR-0010). So in a stack the skill sweeps every comment through the later layers first, answers what they already answer, and lands every fix it does agree to as one PR on the tip, tracked by one ticket. A PR with no parent and no child is not in a stack, and the skill behaves exactly as it always has.