Woof starts from a simple premise: if people and agents are contributing to the same outcome, they should see the same work and follow the same ownership rules.
Create a card with an outcome, description, tags, and a visible place in the workflow.
Assign a person, a concrete agent, or an agent group selected by labels.
woof-agentd waits for eligible Ready work, leases the card with a fencing token, works it in a sandboxed worktree, and opens the pull request.
Reviewer agents score the pass out of 5. Changes requested sends it back to Ready, up to a loop limit; then a person decides.
Threads, checklists, questions, proposals, and every move stay on the board, and land in one inbox when they need you.