JakeB-5

Capture For Agents — AI skill for Claude Code

AI community

여기요!! (Yeogiyo.

How to install Capture For Agents

This entry records only its repository, not the path inside it, so there is no exact command to give. Open JakeB-5/capture-for-agents and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Capture For Agents does

여기요!! (Yeogiyo!!) — macOS screenshot capture + annotation for AI coding agents. Point at the screen: the fix goes right HERE!! (CapNote v1, Tauri v2)

Alternatives in AI

  • Desktop Cc Gui — Multi-engine AI coding desktop client (Tauri) 4.4k ★
  • Peekaboo — Peekaboo is a macOS CLI & optional MCP server that enables AI agents to capture screenshots of applications, o 3.2k ★
  • Open Science — Open Science Desktop — local-first, model-agnostic AI research workbench for macOS, Windows & Linux 1.5k ★

README

Yeogiyo!! (여기요!!)

**English** | [한국어](README.ko.md)

A macOS screenshot capture + annotation tool that delivers visual feedback to AI coding agents — precisely.

*Yeogiyo!!* (여기요!!) is Korean for calling out **"over HERE!!"** — the feeling of pointing a finger at the screen: *"the fix goes right here!!"* (formerly *Capture for Agents*)

**Status: ✅ Phase 3 complete (2026-07-03) — distributable build.** The full loop (capture → annotate → ⌘V) is verified end-to-end on a real device, including history, settings, and a release DMG (ad-hoc signed). Notarization (no Apple Developer account yet) is the only deferred item; multi-monitor coordinates are field-verified on an external 4K display. The detailed work plan lives in [docs/plan.html](docs/plan.html) (Korean).


Why

You implemented a Figma design with Claude Code, and the result is subtly off: a card is missing its 1px border, a margin that should be 24px is 12px, an icon renders one size too small.

How do you tell the agent?

  • Throw a raw screenshot at it and the agent doesn't know where you mean. It can tell something is wrong somewhere, but pinpointing which card, which edge from coordinate intuition alone is unreliable.
  • Describe it in words only, and you're writing paragraphs to disambiguate "the card border is missing."

Visual information and semantic information travel separately — that's the problem.

Yeogiyo!! solves it with **numbered labels + notes + a dual channel**:

  • Drop numbered markers, rectangles, and arrows onto the captured image (visual channel),
  • attach a free-text note to each number (semantic channel),
  • and put a single CapNote v1 text block on the clipboard that joins both channels by number.

The agent reads the image path from the text, opens the file directly, and matches the burned-in numbers `[1] [2] [3]` against the coordinates + notes in the text. "Exactly where, exactly what to fix" arrives in one ⌘V.

How it works

  1. ⌥⇧C — global shortcut starts