# Provider adapter templates

These are request-body templates, not executable requests. They contain no API key and do not authorize a network call.

- `OpenAI API Skills`: review the published OpenAI / Codex Skill ZIP, upload it through `/v1/skills`, retain the returned `skill_id`, and attach the corresponding Skill reference under the hosted shell environment in the Responses request. Upload alone does not mean the Skill was used.
- `openai-responses.json`: use the Responses API top-level `instructions` field.
- `openai-chat-completions.json`: use a `developer` message on models that document that role.
- `deepseek-chat.json`: use the provider-documented `system` role.

Before changing an integration, inspect the provider's current official documentation and the application's existing request path. Show the proposed configuration diff, model identifier, data retention setting, test payload, expected cost, and rollback. Obtain explicit user authorization before editing configuration or sending a test request.

OpenAI currently recommends Responses for new projects: https://developers.openai.com/api/docs/guides/migrate-to-responses
OpenAI API Skills guide: https://developers.openai.com/api/docs/guides/tools-skills
