Design loop engineering workflows for coding agents
Use Loop Engineering to scaffold, audit, and operate repeatable coding-agent loops with explicit state, budget, readiness, and tool-specific patterns.
npx skills add agentskillexchange/skills --skill design-loop-engineering-workflows-for-coding-agents
Use this skill when an operator wants to turn one-off prompting into a repeatable coding-agent workflow. The agent should inspect the repository goal, choose an appropriate loop pattern, scaffold the loop files, run readiness and cost checks, and produce reviewable next commands for Claude Code, Codex, OpenCode, Grok, Cursor, or another supported coding agent. The workflow is strongest for teams that need durable state, budget control, worktree isolation, and a written loop contract before letting agents run repeatedly.\n\nThe operator invokes this instead of chatting with the product normally when the task needs ongoing control: daily triage, recurring fixes, multi-step repository maintenance, or agent runs that must be audited before execution. The boundary is the loop design and operating system around a coding agent. It is not a general prompt guide or a directory card; the output should be concrete loop files, readiness scores, cost estimates, MCP lookup configuration where useful, and a short runbook that keeps each invocation reviewable.