Type: url
Original: https://developers.openai.com/blog/skills-shell-tips
Status: partial
Summary
An OpenAI developer blog post (February 11, 2026) presenting practical patterns for combining three agentic primitives in the Responses API — Skills aligned with the Agent Skills open standard, a hosted shell tool running in OpenAI-managed containers, and server-side compaction — for long-running agents. It gives ten tips, among them writing skill descriptions like routing logic, adding negative examples, keeping templates inside skills, restricting network access with two-layer allowlists and domain secrets, and using /mnt/data as the artifact handoff boundary, and three build patterns illustrated partly with results reported by the early skills customer Glean.