continuous-agent-loop

To run a controlled autonomous agent loop, apply quality gates, evaluation checkpoints, and persistence controls. This prevents churn, stalls, and cost drift during long-running execution cycles.

To recover a stalled autonomous loop, freeze the current scope and audit the execution harness. This addresses common failures like repeated retries, merge stalls, and cost drift.

For CI-gated delivery, select a strict CI/PR control loop pattern. Alternatively, use RFC-driven decomposition, exploratory parallel generation, or a default sequential flow based on your specific task.

Evaluation checkpoints are needed throughout an autonomous development cycle to keep work measurable and aligned with quality gates. They prevent uncontrolled churn and ensure progress remains recoverable.

Cost drift and repeated retries in an agent workflow occur due to missing quality gates and persistence controls. Freezing scope and auditing the harness helps realign the execution loop.

Yes, continuous-agent-loop supports RFC-driven decomposition. It recommends combining it with quality gates, eval loops, and session persistence to compose robust production workflows.