rat-ultraloop

What problem does it solve?

Automates long-running implement-review-improve cycles so teams can run unattended RTL implementation passes while preserving design-freeze integrity and applying automated fixes discovered during review.

Core Features & Use Cases

• Dispatches a target skill (e.g., rtl-p4-block-parallel) repeatedly to continue implementation across cycles.
• Runs a read-only ultraloop-reviewer to assess RTL quality, extract prioritized improvements, and apply must-fix changes automatically.
• Executes contract tests per block, verifies design-freeze via a frozen-hash snapshot, and performs fail-closed stash + halt on freeze violations.
• Supports state persistence (.rat/state/ultraloop-state.json), 30-minute auto-continue via stop-gate.sh escalation, and structured return reports for user review.
• Use case: run unattended block-parallel Phase 4 to merge blocks iteratively while enforcing safety and automated remediation.

Quick Start

Run the ultraloop targeting a block-parallel skill with the command /rtl-agent-team:rat-ultraloop rtl-p4-block-parallel.