← 返回
未分类 中文

Codex: Finish the Task

Autonomous Codex run-to-completion workflow for low quota/cota situations. Use when the user asks for an uninterrupted agent run, says "nao me peca confirmac...
Autonomous Codex run-to-completion workflow for low quota/cota situations. Use when the user asks for an uninterrupted agent run, says "nao me peca confirmac...
vyctorbrzezowski vyctorbrzezowski 来源
未分类 clawhub v1.0.0 1 版本 100000 Key: 无需
★ 0
Stars
📥 191
下载
💾 0
安装
1
版本
#latest

概述

Codex Run to Completion

Operating Mode

Execute the requested task end to end. Treat the user's latest prompt as permission to make routine choices, run commands, edit files, wait for output, and verify results without asking for avoidable confirmations.

Rules

  • Do not stop to ask for confirmation for reversible, routine, or clearly implied steps.
  • Make the smallest reasonable assumption when context is missing; keep moving and mention the assumption in the final summary if it mattered.
  • Self-manage long waits in the terminal. Use bounded polling, sleep, repeated output reads, status checks, or log inspection as needed.
  • Do not end the turn while a command, server, test, build, job, or browser verification needed for the task is still running.
  • Prefer targeted reads and focused commands. Low quota/cota means conserve context and tool calls, not stop early.
  • Keep progress updates short and only use them to report actual movement, blockers, or waiting state.
  • Finish with the concrete outcome: what changed, what ran, what passed or failed, and any exact remaining blocker.

Real Blockers

Stop only when continuing would be unsafe, destructive, outside the user's request, blocked by missing credentials, blocked by required external approval, or impossible with the available tools.

When blocked, do not ask a broad question. State the blocker, the last completed step, the next command or decision needed, and the safest partial result.

Confirmation Policy

Proceed without asking when the action is normal for the task:

  • reading files, searching the repo, inspecting logs, checking status
  • installing local dependencies already required by the project
  • running tests, builds, typechecks, linters, smoke tests, or dev servers
  • editing files directly related to the request
  • waiting for output or polling until a command finishes
  • choosing a conservative default when multiple equivalent paths exist

Ask or stop only for actions with real external or destructive consequences, such as deleting user data, force-resetting work, publishing public comments, spending money, rotating credentials, or changing production systems when the user did not request it.

版本历史

共 1 个版本

  • v1.0.0 当前
    2026-05-29 14:14

安全检测

腾讯云安全 (Keen)

队列中

腾讯云安全 (Sanbu)

队列中

🔗 相关推荐

ai-agent

self-improving agent

pskoett
记录自身发现以实现自我改进的技能
★ 4,165 📥 939,228
knowledge-management

Session Brief

vyctorbrzezowski
生成当前会话的紧凑完整上下文,直接在聊天中输出,适用于用户需要获取对话摘要以供参考的场景。
★ 0 📥 311
ai-agent

Find Skills

root
帮助用户发现和安装智能体技能,当用户询问如「如何做X」、「找X的技能」、「有能做...的吗」等问题时
★ 1,519 📥 577,259