OpenAI's GPT-6 family guide offers prompting guidance for autonomous task progress. The Chinese prompt below is an editorial rewrite, and its use with Sol still needs to be evaluated against the target task.
Suitable tasks: Coding, research, or tool workflows with a clear goal that require continued execution; situations where the model should use context to fill in routine details first, then ask about ambiguities that could materially change the result.
Unsuitable tasks: Tasks that require unconditional execution, ignoring authorization boundaries, or accepting no questions from the model about important ambiguities.
Applicable model version: The source is aimed at the GPT-6 model family. Its behavioral description originated with GPT-6 Astra; use with gpt-6-sol still requires evaluation against the local task and does not represent Sol-specific validation.
Applicable client, agent, or API: API applications, Codex, or other agents that can accept system or developer instructions. The prompt text itself does not provide tools, permissions, or an execution environment.
Recommended reasoning level and parameters: The source does not specify a reasoning level or sampling parameters for this prompt. Use the application's baseline configuration and record the actual settings.
The following Chinese prompt is an editorial adaptation of the official prompting intent, not an official verbatim prompt:
Use the user's request, conversation context, and project instructions to determine the goal the user wants to achieve and the boundaries of the current task. Continue work that is clear and authorized, completing steps supported by the information already available, and incorporate any new requirements from the user along the way.
When routine details are missing, proceed using reasonable assumptions that fit the context, and state key assumptions in the result. Ask a focused question only when missing information could materially change the result or the boundaries of the action.
During execution, respect the user's instructions and the scope of their authorization. For irreversible actions, external publication, or other explicit boundaries, follow the application's established approval requirements first. When finished, briefly report the result, verification performed, and any unresolved items.Short excerpt from the official source: “bias towards action and carry the user's intended task to completion”.
Select a representative set of tasks for Sol, covering continued execution when information is sufficient, important ambiguities, mid-task user corrections, and action boundaries.
Fix the model version, reasoning level, tools, permissions, and task inputs. Compare task completion rate, incorrect assumptions, unnecessary questions, boundary violations, and human takeover with and without this prompt.
Check whether important ambiguities trigger a question, routine omissions are handled reasonably, and new user requirements are incorporated into subsequent work.
Record failure cases and adjust the prompt. Do not treat Astra's behavioral description or family-level prompting recommendations as validated results for Sol.
The official page says these prompts can be used as a starting point for the GPT-6 family. It explicitly states that the prompts target behavior observed in GPT-6 Astra and should be evaluated for the selected model and workload.
The Chinese template is an editorial rewrite of the autonomous-progress prompt and nearby explanatory text. Its guidance on stating assumptions, checking authorization, and reporting results is for application-level use and is not verbatim source text.
This prompt cannot replace an application's permission controls, tool implementation, approval process, or task acceptance criteria.
GPT-6 Sol