Type: url
Original: https://developers.openai.com/blog/skills-agents-sdk
Status: partial
Summary
An OpenAI developer blog post describing how the maintainers of the OpenAI Agents SDK’s Python and TypeScript repositories use Codex with repo-local skills, AGENTS.md and the Codex GitHub Action to turn recurring work — verification, docs sync, example runs, release review and PR handoff — into repeatable workflows, alongside a rise in merged PRs over the period described. Its practical lessons are to make skill use mandatory through short if/then rules in AGENTS.md, to treat a skill’s description field as routing metadata, to put deterministic shell work in scripts while leaving interpretation to the model, and to keep human review for decisions between several valid options rather than for routine correctness.