SHIPPED · v1.19.0
Scripted and agent-recorded tutorials
New
Open a showcase spec — Record ▸ Scripted recording takes a YAML file that says what to narrate, what to click and what must be true afterwards, and plays it against the window you pick
Narration measured before anything is pressed — every line is spoken first and each step is held for its words, so voice and picture are in step by construction; a machine with no speech engine times the steps instead and says so
Finds the control, then says how — by test id, text or role from the app under test, by a reference picture, or by a description; each step records which one found it
Checks as it goes — an expectation that fails stops the run on that step and names it, and what was recorded is still a normal take
Arrives edited — the take opens with its voice-over, camera moves, callouts, captions and masks already placed, and with the review still to do
Say record this feature demo <path> — Claude Code, Codex, Grok, Gemini CLI, Cursor and the other connected agents can record a tutorial from a sentence; a spec records, a folder of specs records back to back, and a feature's source folder gets a spec drafted from the controls already in that code. Add your own guidelines in the same sentence
Connect every agent here — Settings ▸ Agent access lists ten coding agents (Claude Code, Claude Desktop, Cursor, Codex, Grok, Gemini CLI, Qwen Code, Windsurf, Kiro and VS Code), shows whether each is connected, and connects every one installed on this machine with one press — never creating a config for an agent that isn't there
You approve every run — runs can be Off, Ask or Allow; an agent's request shows what it will record, what it will launch and how long it will take, and only the main window can answer it. The activity log records decisions, never what was sent
Record ▸ Tutorial harness — give a brief and a source folder, pick a target and an agent, and watch the planned steps run; an Explore session lets the agent click through an app it has no spec for, with every session approved by you, and drafts a spec from what it did
A review queue for handed-back takes — Handoff review lists each take under Needs review with its step guide and what the agent did; Accept stays locked until you have watched it in the studio, and Reject deletes nothing
Agents can only record — no tool can review, export or delete anything; that stays yours
Improved
Autopilot on Windows — Autopilot, scripted recording and the tutorial harness press keys and click through Windows' own input, so all three now run on Windows as well as macOS
Captions read as phrases — subtitles now show a whole phrase at a time, cut at sentence ends, commas and pauses, appearing a beat before the voice and holding through short pauses, instead of a line that slid along word by word; the .srt and .vtt sidecars use the same phrases
Captions land on their words — a narrated take times each word from the voice's own pauses, so captions no longer drift up to a second ahead of what is being said
Fixed
Fixed Windows and Linux recordings running two seconds late — captions, camera moves, click ripples, callouts and timed masks drew ahead of the picture; takes now start at their first real frame, and older takes are corrected when opened

Settings, Agent access: the Coding agents list with Claude Code, Codex, Grok and Gemini CLI connected with the skill installed, Cursor and Kiro connected, and the Activity log of agent calls below 
The An agent wants to record a spec dialog over the Tutorial harness screen, naming the client, the spec, the launched window, a 50-second plan and the exact command it will run, with Decline and Record it 
Handoff review with four unwatched agent takes in the review queue, the Review gate is shut banner with Open in Studio, the Before you watch checks and the step guide, and Export MP4 locked until review 
The Scripted recording screen with an eight-step spec loaded, listing Say, Click and Point at steps with their narration, locators and zoom, beside the Before you press Start notes