Context packs
The folder every group and walkthrough becomes, file by file, and how to use it with Claude Code.
Everything Startwell Capture makes is meant to be read by an AI agent first and a person second. So every capture, in its session, is also written to disk as a context pack: a plain folder of Markdown, images and JSON that any agent can read with no plug-in. A voice note or meeting reads like a walkthrough, with where you were, what you clicked and what was on screen woven through the words. A screenshot, note or link says what you did in the two minutes before it. Switch this off in Settings.
Layout
README.md the library: one line per topic, then what changed lately DAILY.md what changed, day by day HANDOFF.md a ready-made prompt over recent recordings Demo Shop/ one folder per topic; Unsorted/ for the rest README.md the topic: its brief, then its packs and loose captures TOPIC.md the brief alone HANDOFF.md a ready-made prompt over this topic manifest.json 2026-11-03 1402 Demo card declined at checkout/ README.md start here transcript.md steps.md walkthroughs steps/001.png … 005.png walkthroughs; voice notes and meetings when the screen changed issues.md when something was raised captures/ screenshots and audio from the session video.mp4 walkthroughs with video on audio.m4a walkthrough narration manifest.json
Folders are named YYYY-MM-DD HHMM title in local time. The library follows the database: move a capture to another topic and its pack is rewritten. Packs are generated output, so edits inside them are overwritten; put notes in the app instead. Change the folder in Settings, Library.
Every file
README.md
The entry point. It opens with a line telling an agent what the folder is, then always the same sections:
- What this is: kind, when, topic, contents, and roughly how many tokens it takes to read.
- What I did: a numbered list of what happened on screen, from the activity lines. Read it as repro steps.
- People: who spoke, and for how long.
- Chapters: one per page in a walkthrough; in a session, a new one after five quiet minutes or a Focus change.
- Steps: each screenshot with its time, page and what was said.
- Captures: notes, screenshots (with the text read from them), links and recordings.
- Files: what else is in the folder.
> **For AI agents:** this folder is a Startwell Capture context pack. Read this > README first, then open the files it links (paths are relative to this folder); > `manifest.json` has the same data as JSON. # Demo: card declined at checkout ## What this is - **Kind:** Walkthrough — screens captured as someone used an app, with their narration - **When:** Tue, 3 Nov 2026, 14:02 to 14:03 (1 min), local time - **Topic:** Demo Shop - **Contents:** 5 steps, narration (1 min), 1 note - **Size:** ~1.7k tokens to read README, transcript and steps ## What I did 1. Opened Google Chrome at demoshop.example/basket · 00:00 → [step 1](steps/001.png) 2. Clicked menu "Quantity" in "Blue mug" · 00:09 → [step 2](steps/002.png) 3. Clicked link "Go to checkout" · 00:16 4. Went to demoshop.example/checkout — "Checkout — Demo Shop" · 00:17 → [step 3](steps/003.png) 5. Clicked into field "Card number" · 00:36 → [step 4](steps/004.png) 6. Clicked "Pay now" · 00:45 7. Went to demoshop.example/checkout/payment — "Payment failed — Demo Shop" · 00:51 → [step 5](steps/005.png)
transcript.md
Every word, one block per speaker turn: the name, the time from the start of the pack, then the words. Activity lines sit between the turns in italics, at the moment they happened.
_00:00 · Opened Google Chrome — "Your basket — Demo Shop" (demoshop.example/basket)_ **Jo** · 00:01 Okay, this is Demo Shop. I'm on the basket page with one blue mug. _00:09 · Clicked menu "Quantity" in "Blue mug"_ **Jo** · 00:09 I'll make it two.
steps.md and steps/
Walkthroughs only. One entry per screenshot, in order, grouped by chapter: the image, the app, the window, the address, why it was taken, and what was said from about ten seconds before it until the next step.
### Step 4 · 00:45 · Checkout — Demo Shop  - **App:** Google Chrome - **Window:** Checkout — Demo Shop - **URL:** https://demoshop.example/checkout - **Captured:** click (just before) - **Said:** And pay now.
Images are numbered 001.png upwards in timeline order. When something was blurred, the blurred copy is the one in the pack.
captures/
Screenshots and audio from the session’s other captures, named by time and kind, for example 140315-screenshot.png.
video.mp4 and audio.m4a
The walkthrough’s screen recording, if video was on, and its narration audio.
manifest.json
The same data as JSON for tools: speakers, chapters, transcript turns (with the raw text when tidying changed it), steps, captures and activity, each with times in milliseconds from the start. The format is versioned ("version": 1).
Using a pack with Claude Code
When a meeting or walkthrough finishes, its folder path is on your clipboard. For anything else, ⌘⇧Y copies the last capture’s path and ⌥⌘⇧Y the current session’s pack (Win+Shift+Y and Win+Alt+Shift+Y on Windows, Super+Shift+Y and Super+Alt+Shift+Y on Linux). Then:
> Read "~/Startwell Capture/Demo Shop/2026-11-03 1402 Demo card declined at checkout/README.md" and fix the bug it describes.
The README tells the agent what to open next. For a whole topic, point it at the topic folder’s README, or connect the topic so the agent can list and read packs itself.