What problem does it solve? Running multiple Trellis tasks one by one requires constant manual confirmation, route selection, and status tracking. This Skill automates the full loop: preparing tasks, resolving open questions, implementing, checking, and committing locally, all driven by the authoritative auto_loop.py runner state. ## Core Features & Use Cases - Unattended Batch Runs: Start a commit-only run over an explicit task list with dependency declarations, check depth, and implement/check route modes. - Runner-Driven Actions: Execute prepare and running actions (classify dirty baselines, resolve open questions, run implement/check/fix, commit-only) and record each result precisely before advancing with next. - Bounded Autonomous Decisions: Log low-risk local decisions via the decide command with evidence and risk levels, while blocking irreversible or scope-expanding changes. - Recovery Protocols: Handle retryable artifact-drift and recovery diagnostics with a three-attempt correction budget using the reconcile flow in references/artifact-recovery.md. - Use Case: A developer with five planned Trellis tasks starts an auto-loop run; the Skill prepares all tasks, resolves open questions with the user once, then implements, checks, and locally commits each task without further prompts, ending with a pending-archive handoff report. ## Quick Start Ask the AI to start a Trellis auto-loop run over your listed tasks and let it drive auto_loop.py through prepare, implement, check, and local commit until the queue finishes.
