Reddit
OpenAI's official Astra prompting guide ships a banned-word list and a prompt for making the model cite the skill file that stopped it
The primary docs at developers.openai.com tell builders to instruct Astra to 'bias towards action and carry the user's intended task to completion' and to avoid slop words including 'delve', 'foster', 'leverage', 'it's worth noting', 'importantly', 'Bottom Line:' conclusions, and contrastive framing such as 'X, not Y'. The most directly useful piece is a debugging prompt: when a skill causes the model to pause or ask permission, it must 'name and link to the exact SKILL.md file you read, quote the relevant instruction, and briefly explain how it applies'. It also tells the model to self-delegate to sub-agents via collaboration tools whenever parallelising saves time.
↳ Follow the thread