Run autonomous coding tasks through Zeroshot verifier loops
Use Zeroshot when a coding issue or spec needs an executor agent and an independent verifier loop before the change is accepted.
npx skills add agentskillexchange/skills --skill run-autonomous-coding-tasks-through-zeroshot-verifier-loops
Use Zeroshot when a repository task should be implemented by one coding agent and judged by a separate verifier instead of letting the same agent declare its own work complete. The workflow runs a task from a GitHub issue, markdown spec, URL, key, or inline prompt, executes the change in an isolated worktree or container, records progress in a crash-safe SQLite ledger, and repeats implementation and verification until the verifier approves or returns a reproducible failure.
What this skill actually does
Invoke this instead of manually prompting Claude Code, Codex, Gemini CLI, or OpenCode when the task is large enough to justify unattended iteration and independent review. It is most useful for bug fixes, feature work, issue queues, and production changes where the operator wants test-backed approval rather than a single-agent confidence claim.
The boundary is the executor-verifier software engineering loop. The skill should run documented Zeroshot commands, choose an appropriate provider CLI/backend, monitor status/logs, and hand back approved changes or concrete failures. It is not a generic multi-agent framework listing, and it does not replace the underlying provider CLIs.