Guide

• First run: the interview: models, harness, limits, project, you.
• Commands: request, qa/review/check, status, steer, resume, setup, selftest, done.
• Single stages: QA or review on a PR you did not build with Gru.
• Task cards: objective, definition of done, size.
• Models, roles and harnesses: strong/mid/weak, the role matrix, tether fallback, limits.
• How a card runs: plan, tests, build, gate, qa and review, replan, hand-back.
• Sandbox: how minions are contained.
• Budget and calibration: caps, cost, learning the project.
• Logs and store: the cross-session log.
• Feedback to Gru: how Gru learns from your runs.
• Role agents: planner, swe, qa, reviewer, team-leader.