see

# See — give Claude eyes on recent screenshots The user often takes a screenshot and then wants Claude to look at it. This skill turns that into a one-step interaction: `/see` finds the screenshot, opens it, and Claude can describe what's on screen. ## How invocation maps to behavior - `/see` (no argument) → list the 5–10 most recent screenshots with their filenames and how long ago they were taken. Ask which one to open, OR if the most recent is clearly the one (taken in the last minute or two), just open it directly. - `/see N` where N is an integer → open the Nth most recent screenshot. `/see 1` is the most recent, `/see 2` is the one before that, etc. - `/see <text>` → if it's not a number, treat it as a filename substring and open the most recent match. (Optional, light-touch.) The user may also ask for this in natural language ("look at my last screenshot", "what was in that snag I just took", "see the screenshot from a minute ago"). Treat those the same as `/see 1`. ## Configuration Config lives at `~/.see-skill/config.json`: ```json { "screenshot_dir": "/absolute/path/to/screenshots" } ``` This single file is what makes the skill portable — once it exists, the same skill works identically in Claude Code, Cowork, or anywhere else. ## First-run setup If `~/.see-skill/config.json` does not exist: 1. Try to auto-detect the macOS system screenshot location: ```bash defaults read com.apple.screencapture location 2>/dev/null ``` 2. If that returns nothing useful, ask the user where their screenshots live. (Snagit users often save to `~/Pictures/Snagit`; the macOS default is `~/Desktop`.) 3. Save it: ```bash mkdir -p ~/.see-skill printf '{ "screenshot_dir": "%s" }\n' "$DIR" > ~/.see-skill/config.json ``` After setup, confirm: "Got it — I'll look in $DIR. From now on, just say `/see`." ## The actual workflow each time 1. Read the config: ```bash DIR=$(python3 -c "import json; print(json.load(open('$HOME/.see-skill/config.json'))['screenshot_dir'])") ``` If the file is missing, do the first-run setup above. 2. List recent screenshots (or pick one). Use the bundled helper: ```bash bash <skill-path>/scripts/recent.sh # list 10 most recent bash <skill-path>/scripts/recent.sh 3 # print absolute path of 3rd most recent ``` 3. Open the chosen file with the Read tool. The Read tool renders images natively, so you'll see the screenshot inline and can describe what's in it. 4. Then respond to whatever the user actually wanted — describe the screenshot, answer a question about it, draft a reply to what's on screen, etc. ## Notes on edge cases - **Folder missing or empty**: tell the user, offer to reconfigure (`rm ~/.see-skill/config.json` and re-run setup). - **Snagit-style filenames** are descriptive ("2026-05-06_login_error.png"). When listing, show the filenames — the user often recognizes the one they want by name. - **Recognized image extensions**: png, jpg, jpeg, gif, tiff, heic, plus Snagit's `.snag` / `.snagx`. The helper script already filters to these. - **Snagit `.snagx` files are zip archives**, not images. When you pick one (e.g. `recent.sh 1`), the script transparently extracts the embedded full-resolution PNG into `~/.see-skill/cache/` and returns the path to that PNG — that's what you pass to Read. Listing (no argument) shows the original `.snagx` filenames so the user can recognize them. - **Don't hardcode the directory** in the skill itself — always read it from the config file. That's what keeps the skill portable across machines.