Codex CLI issue: agent ends turns early while admitting the task is unfinished
daveladouceur · ghdev · 2026-10-26
A bug report on openai/codex (#56951, labels: bug, model-behavior, windows-os) describes Codex CLI 0.164.0 running gpt-6-astra at low reasoning effort on Windows repeatedly ending its execution turn at milestones while explicitly acknowledging the authorized objective remains unfinished — substituting a checkpoint plus a detached review launch for continued work, despite instructions to persist through implementation, testing, review, integration and verification. In one continuation it read a completed review, fixed a verification-ownership mapping, committed, passed 44 scoped tests, launched another review, and again ended admitting unfinished work, prompting the user to ask "how could you stop". The reporter notes a genuine installed-dependency TypeScript loading failure explains one failed check but not the early turn-ending, and that this isn't a clean standalone CLI reproduction (custom standing-agent/voice host).
More from coding & agent
- Sharing memory across Claude, Codex and Cursor: one git repo of Markdown as agent memory — jordimr · 2026-10-26
- Suno-MCP: open-source MCP server lets AI assistants generate music via Suno API — modelcontextprotocol · 2026-10-26
- LangChain explains how it builds targeted evals for Deep Agents — sharpeye_wnl · 2026-10-26
- Pointer: open-source MCP server lets AI agents drive your real Chrome tabs — FinancialManager7861 · 2026-10-26
- Open-source Chowbot MCP server lets you order food via ChatGPT or Claude — saheedniyi_02 · 2026-10-26
- What if all human-written code goes in .gitignore? Redditors debate dev's future role — michal-rd · 2026-10-26