loop-engineeringlisted

# Loop Engineering Build and operate the infrastructure that runs autonomous agent loops — the mechanism layer that schedules, monitors, scores, and recovers loops. This is the engineering counterpart to loopy (which designs loop content and boundaries). ## When to use - Building a new autonomous agent loop from scratch - Adding scheduling, monitoring, or recovery to an existing loop - Auditing loop infrastructure for reliability gaps - Setting up loop governance (rate limits, budgets, approval gates) - Debugging a loop that stalls, spins, or silently fails ## Core concepts ### Loop anatomy Every loop has: - **Trigger**: What starts it (cron, webhook, event, manual) - **Body**: The work it does (triage → act → verify) - **Gate**: What stops it (budget, time, approval, success signal) - **Recovery**: What happens on failure (retry, escalate, pause, alert) ### Loop health scoring Score loops on: - **Completion rate**: % of runs that reach a terminal state - **Budget adherence**: % of runs within token/cost budget - **Action rate**: % of runs that produce a meaningful action (not just "no-op") - **Recovery rate**: % of failures that self-recover without human intervention ### Governance patterns - **Budget caps**: Hard limits on tokens, cost, or wall-clock time per run - **Approval gates**: Human-in-the-loop for high-impact actions - **Rate limits**: Max runs per hour/day, cooldown between runs - **Audit trail**: Every run logs its trigger, actions, and outcome ## MCP