v0.36.1: Hand your agent a screen recording, and Sonnet 5.5 and GPT-6.1 Sol take over as defaults
v0.36.1 is a small release with two additions: your agent can now watch a screen recording, and newer models are the defaults.
Hand your agent a screen recording
Coding agents can read images but not video, so a bug-report clip or a motion mockup used to be a dead end. The new /aitask-screen-recording skill turns a recording into contact sheets, key frames and a timeline that the agent can read. For animations, it also measures duration and easing, so “match this transition” gets real numbers instead of a guess. You need ffmpeg installed, and nothing is uploaded or written into your repo.
Sonnet 5.5 and GPT-6.1 Sol take over as defaults
Claude Sonnet 5.5 is registered, along with a 1M-context variant. It is now the default wherever Sonnet 5 was, including explain, QA, batch review, work reports and the brainstorm operations. On the Codex side, GPT-6.1 Sol replaces GPT-6 Sol for shadow and discuss agents. The older models remain registered if you want to pin them.
Full changelog: v0.36.1 on GitHub