A Claude Code skill for SlimSnap. Lets Claude Code automatically read structured JSON captures of your screen, so you can say "fix this layout" instead of pasting a screenshot every turn.
About 700 tokens per capture vs the API per-image cap of 1,568 tokens on Sonnet/Haiku (up to 4,784 on Opus 4.7+ per Anthropic's vision docs). Structured bounding boxes, extracted colors, OCR text, and your annotations.
Globally, available in every project:
mkdir -p ~/.claude/skills/slimsnap
curl -L https://raw.githubusercontent.com/bickov/slimsnap-skill/main/SKILL.md \
-o ~/.claude/skills/slimsnap/SKILL.mdOr scoped to a single project:
mkdir -p .claude/skills/slimsnap
curl -L https://raw.githubusercontent.com/bickov/slimsnap-skill/main/SKILL.md \
-o .claude/skills/slimsnap/SKILL.md- Launch SlimSnap at least once so it publishes its config to
~/.slimsnap/config.json. - Capture a screenshot with SlimSnap (
⌘⇧S). - Annotate what matters: arrows, rectangles, callouts.
- Hit Save JSON (autosave on by default writes straight to your configured folder).
- In Claude Code, say what you want: "fix the broken sign-up layout I just captured."
- Claude Code reads the latest JSON automatically and acts on it.
Pasting a raw screenshot to a coding agent costs hundreds to thousands of vision tokens per turn and the agent re-interprets pixels every time. A SlimSnap JSON costs about 700 tokens, is structured (the agent acts on coordinates), and is reusable across turns without re-paying the cost. Over a long Claude Code session, the difference in context and token spend is real.
The skill is not hardcoded to a specific folder. SlimSnap publishes a tiny config at ~/.slimsnap/config.json naming its current default save folder. The skill reads that file on every invocation and looks in whatever folder you have configured in the SlimSnap Settings window. Change the folder in SlimSnap and the skill follows.
If you prefer keeping captures with a codebase, create a <project>/.slimsnap/ folder and the skill will prefer it when present.
The JSON format is an open MIT spec: https://github.com/bickov/slimsnap-schema
MIT.