What problem does it solve? Coordinating parallel AI agents on one repository without collisions, unverified merges, or lost context is hard. This Skill encodes the architect's seat for a Supertaskr project: it derives what work is startable, dispatches lanes through a single eight-step arm, enforces blind two-spawn verification, and governs the verdict, merge, checkpoint, and push so every landing leaves the same auditable files. ## Core Features & Use Cases - Derived dispatch view: Computes live lanes and startable cards from git worktrees and the board before stamping, never from stale status fields. - One-arm dispatch: Runs the eight-step dispatch arm (stamp, cut lane, preflight, fence manifest, bench, brief, port derivation) that refuses at the first failing step, with a --dry-run plan mode. - Blind verification protocol: Spawns the verifier in two phases where the attack plan is written before the diff exists, with hashed attack sets cited in the verdict. - Self-checking tooling: Ships golden-check.mjs and host-command-check.mjs, zero-dependency Node scripts that compare landed artifacts field-by-field against a golden contract and verify every command is transcribed from project conventions. - Use Case: You ask what to work on next in a Supertaskr repo; the skill derives the dispatch view, triages suggestions, audits the card's claims, dispatches the lane, and lands it with a verdict. ## Quick Start Open a Supertaskr project and ask the agent to dispatch the next startable card using the supertaskr-seat skill.
