Using skills to accelerate OSS maintenance

Description and use

Automating OpenAI SDK maintenance with Codex skills, AGENTS.md, and GitHub Actions increased PR merge rate by 44 % over 90 days, ensuring consistent code, docs, and release checks.

Problem

Maintain OpenAI Agents SDK repos efficiently while ensuring code quality, documentation, and release integrity.

Outcome

Automated workflows increased PR merge rate by 44% (316→457) over 90 days, reduced manual effort, and ensured consistent verification, documentation, and release processes across both SDK repos.

Workflow

• Define repository policy in AGENTS.md
• Create repo‑local skills in .agents/skills/
• Add optional scripts/references inside skills
• Configure Codex GitHub Action for CI
• Use skills for recurring tasks (e.g., code‑change‑verification, docs‑sync, examples‑auto‑run, final‑release‑review, implementation‑strategy, openai‑knowledge, pr‑draft‑summary, test‑coverage‑improver)
• Enforce mandatory skill usage via AGENTS.md rules (e.g., call $implementation‑strategy before runtime changes, $code‑change‑verification for code/build changes, $changeset‑validation for JS package changes, $openai‑knowledge for API work, $pr‑draft‑summary before handoff)
• Run verification stacks defined in skills (e.g., make format lint typecheck tests for Python, pnpm i build … for JS)
• Automate release validation and upgrade workflows (e.g., changeset‑validation, pnpm‑upgrade)

Evidence

Repo-local skills, AGENTS.md, and GitHub Actions let us turn recurring engineering work, such as verification, release preparation, integration testing for examples, and PR review, into repeatable workflows.

Source

Between December 1, 2025 and February 28, 2026, the two repos merged 457 PRs, up from 316 in the previous three months from September 1, 2025 through November 30, 2025

Source

The Python repo is the simpler baseline: code‑change‑verification runs the required formatting, lint, type‑checking, and test stack when code or build behavior changes.

Source