OpenAI recommends developers streamline prompts and reduce guardrails for GPT-6 Astra to enhance performance, according to a recent blog post. The company suggests tying instructions more tightly to specific tasks and clearly defining when a job is complete. This advice aligns with earlier guidance from OpenAI on transitioning between models.
Overly long skill descriptions and rigid approval rules can hinder GPT-6 Astra's effectiveness, writes OpenAI's Eric Provencher. He notes that vague skill descriptions lead to incorrect selections, as the model relies on context to choose the right skill for a task. Skills, stored as Markdown files, are used to guide the model's actions.
Provencher advises keeping scope descriptions short and precise to prevent Codex from truncating critical information. A skill for Postgres schema migrations, for example, should only activate when creating or modifying a migration. If a skill covers multiple workflows, its main document should briefly point to supplementary resources to ensure the model reads only what it needs.
Mandatory reading requirements before every change can waste context, Provencher says. Requiring the model to read multiple documents for a simple typo fix is unnecessary, as Astra can determine its needs independently. He recommends selectively pointing to documents like architecture.md, database.md, and deployment.md based on the task at hand.
Explicit permissions can also reduce repeated confirmation requests for safe operations, according to the post. For local tests using throwaway data, AGENTS.md can explicitly allow the agent to run tests, fix errors, and re-run affected tests without further input. Astra may stop earlier than GPT-5.6 Sol, Proven, notes, and defining what "done" means upfront is crucial.
OpenAI recently published detailed prompting tips for GPT-6 Astra, and these recommendations build on that guidance. The company did not specify how to handle complex workflows that may require more oversight, and it remains unclear how Astra will balance autonomy with safety in practice.
Source: thedecoder
