What problem does it solve? Delegating work to cheap CLI model workers often produces unverified, invisible output. This Skill enforces an orchestration discipline where every task runs as a manifest with an executable check, so exit code 0 is the only PASS and every run is watched on the Ringside dashboard. ## Core Features & Use Cases - Manifest-based delegation: Write self-contained worker specs with disjoint file ownership, embedded run commands, and output contracts, then lint and run them with ringer.py. - Executed verification: Every task carries a check command that validates content rather than existence, with failure output injected into a single retry prompt. - Pattern playbook and engine routing: Choose from ready-made kits (review-swarm, fix-swarm, focus-group, bakeoff, research-with-proof) and pick worker engines from the local model scoreboard's first-try pass rates. - Use Case: You need five independent bug fixes across a repo. You write a fix-swarm manifest with isolated git worktrees and patch-export checks, run it under Ringer, watch progress on Ringside, and review the exported patches before merging. ## Quick Start Ask the AI to load the ringer skill and turn your next multi-file task into a linted ringer.py manifest with executable checks, then run it while watching the Ringside page.
