Continuous Agent Loop

Defines continuous autonomous agent loop patterns with quality gates, evaluations and recovery controls.

What it does

This skill documents a canonical loop pattern for continuous autonomous agents, covering loop selection, composition and failure handling. It provides a decision flow for choosing between continuous-pr, rfc-dag, infinite and sequential loops, and recommends a production stack combining RFC decomposition, quality gates, an evaluation loop and session persistence. It also lists common failure modes such as idle spinning, repeated retries and cost drift, with recovery steps including freezing the loop, running an audit, narrowing scope and replaying with acceptance criteria.

When to use it

Use it when designing or operating a continuous autonomous agent loop and you need to pick a loop style or assemble a production loop stack. It is also relevant when a running loop stalls, retries the same failure or drifts in cost and you need a recovery procedure.

Requirements

No scripts are included; it is an instructions-only document. It references other named skills and commands such as continuous-pr, rfc-dag, infinite, sequential, ralphinho-rfc-pipeline, plankton-code-quality, /quality-gate, eval-harness, nanoclaw-repl and /harness-audit, which would need to be available separately.