Image: Scrawl, Kit Langton
Why it mattersDescribing a bug to a coding agent in text loses the thing you were pointing at, so a short recording with marks and narration gives the model what you actually saw.
Describing a bug to a coding agent in text is where most of a session goes. Scrawl, a new open-source macOS tool by Kit Langton, offers a shorter path: press one keyboard shortcut, talk, draw on the screen, press the shortcut again. The tool records the screen, your voice and your marks, then packs the lot into a bundle a coding agent can read.
The project is MIT-licensed and has 125 stars on GitHub. It sits in Langton's own domain, scrawl.kitlangton.com, with the same marketing line the README uses: show an agent a problem instead of describing it.
What a capture produces
A tap of ⌥⇧S starts a screen recording. Hold Shift while the mouse moves and the pointer draws a red marker that fades out on release. Talk at the same time and Apple's on-device SpeechAnalyzer transcribes your words with per-word timings. A second tap of ⌥⇧S finishes the capture. The output is a .scrawl folder containing four things: contact-sheet.png, a grid of annotated keyframes picked from the moments that mattered; frames/, each keyframe as a full-resolution PNG; summary.md, a time-aligned transcript and interaction timeline; and scrawl.json, a structured manifest of the whole thing.
Scrawl picks keyframes from screen changes, mouse clicks, laser points, and spoken cues. According to the README, saying "now", "here", "look", or "watch" at a moment where the view has settled turns that frame into a keyframe. Filler words like "this" or "that" are ignored unless the screen was changing at the same time.
Sending it to the agent
The CLI has a send command that copies contact-sheet.png and summary.md to the clipboard, so paste into any chat with an AI coding assistant drops both in at once. The README also calls out a direct route into OpenCode V2 sessions through --session ses_..., and a route subcommand that decides which session a capture of a particular local dev URL belongs to.
Captures of a cabled iPhone are supported through CoreMediaIO, with the phone's screen shown in a mirror window on the Mac. A separate Swift package called ScrawlKit lets an iOS app embed the recorder, with uploads routed back to the Mac over Tailscale through a headless scrawl serve ingest server. Narration from Hex, if it is running, can substitute for Apple SpeechAnalyzer, picking any model Hex has ready.
The macOS 26 requirement and what it carries
Scrawl runs on macOS 26 only, because Apple SpeechAnalyzer is a macOS 26 framework. First launch asks for three permissions: Screen & System Audio Recording, Microphone, and Input Monitoring. Captures sit in ~/Library/Application Support/Scrawl, and the app trashes video, audio and full-size frames after 7 days by default and whole bundles after 30, with scrawl pin to keep one. Settings holds a longer retention preset for teams that want to keep video around.
A bundle a developer sends to a coding assistant travels through whichever model provider that assistant calls, which means your voice and your screen content leave the machine along with the paste. The README is honest about this: the output is a shareable bundle, nothing more.
Source
The code is at kitlangton/scrawl on GitHub and the project page is at scrawl.kitlangton.com.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.