Continuous Agent Loop

This is the updated canonical pattern for autonomous agent loops that run until they hit quality gates or acceptance criteria. You get four loop strategies (strict PR control, RFC decomposition, parallel exploration, or sequential), plus the recommended production stack combines RFC pipelines, quality gates, eval harnesses, and session persistence. The honest take: this prevents the classic failure modes where agents spin in circles retrying the same broken approach or rack up costs without measurable progress. When things do stall, you freeze the loop, audit with the harness, narrow scope to the failing unit, and replay with explicit criteria. It replaced the old autonomous-loops skill but keeps backward compatibility for one release.

Install to Claude Code

npx -y skills add affaan-m/ecc --skill continuous-agent-loop --agent claude-code