Skip to content

About

A standalone Claude skill: point Claude at a folder of images, get back one structured markdown take-away (summary, action items, one transcribed section per image).

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

8 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

takeaway-md

A standalone Claude skill. Point Claude at a folder of images and get back one structured markdown document: a summary, action items, open questions, and then one captioned, faithfully transcribed section per image.

It is deliberately self-contained. It knows nothing about any particular app, file format, or folder structure — its only inputs are a folder path and the image files inside it.

What it does

  1. Enumerates images in the folder: .png .jpg .jpeg .webp .gif .bmp, plus .heic when readable (unreadable ones are reported and skipped). Non-image files are ignored. Subfolders are not recursed. If there are no images, it says so and writes nothing.
  2. Orders them: if every filename carries a NAME_YYYY-MM-DD_HHMMSS timestamp, it sorts by that and puts the time in each heading. Otherwise it sorts by filename, with modification time as a tie-breaker.
  3. Confirms first if the folder has more than 15 images, since processing them all takes a while and uses significant context.
  4. Transcribes each image faithfully — prose as prose, tables as markdown tables, code as fenced code blocks, UI transcribed. Each image gets its own section plus a one-line italic caption of what it is (an annotated transcript).
  5. Writes incrementally to <folder-name>.take-away.md inside the folder, appending each section as it is produced, so an interruption never loses finished work.
  6. Resumes on a re-run: if the output file already exists, images already in it are skipped. Ask for a regenerate ("start over", "regenerate") to rebuild from scratch.
  7. Fills in the Summary, Action items, and Open questions once every image has been processed. Action items and Open questions are omitted entirely when there are none.

Output shape

---
source: <folder-name>/
images: <n>
generated: <ISO timestamp>
generator: take-away
---

# <folder-name> - take-away

## Summary
...

## Action items
- [ ] ...

## Open questions
- ...

---

## 01 - 15:25:36
*one-line caption of what this shot is*

![](first-image.png)

<rendered content>

## 02 - 15:27:59
...

Headings are ## NN - HH:MM:SS, or ## NN alone when no timestamp is available. The output contains no em dashes — hyphens only.

Install

Copy the skill folder into your skills directory:

# project-scoped
cp -r . <your-project>/.claude/skills/takeaway-md

# or user-scoped (all projects)
cp -r . ~/.claude/skills/takeaway-md

The folder needs SKILL.md and scripts/. tests/, README.md, and LICENSE are not required at runtime.

Requires Python 3.9+ on PATH (python or python3) for the folder scan. Pillow is optional — it is only used to probe whether .heic files are readable.

Use

Point Claude at a folder:

process the screenshots in ./meeting-2026-08-30
turn D:\shots\onboarding-flow into notes
make a take-away from ~/Desktop/whiteboard-photos
OCR this folder: ./receipts

Claude runs the scan, checks the gates (zero images, more than 15 images), writes <folder-name>.take-away.md into the folder, and reports where it landed.

To rebuild an existing take-away from scratch:

regenerate the take-away for ./meeting-2026-08-30

Layout

.
├── SKILL.md                    the skill instructions
├── scripts/
│   └── take_away_scan.py       folder scan: enumerate, order, resume-detect -> JSON plan
├── tests/
│   ├── README.md               how to run the fixture test
│   └── fixtures/               three small images (table, code, plain text)
├── README.md
└── LICENSE                     MIT

License

MIT — see LICENSE.

About

A standalone Claude skill: point Claude at a folder of images, get back one structured markdown take-away (summary, action items, one transcribed section per image).

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages