<< All versions

Skill v1.0.0

currentAutomated scan100/100
zen-open-source/pastelocal/paste
──Details
PublishedOctober 22, 2026 at 10:38 AM
Content Hashsha256:c25fe1300d515914...
Git SHA8d0a5311c515
──Files
Files (1 file, 3.1 KB)
SKILL.md3.1 KBactive
SKILL.md · 51 lines · 3.1 KB

version: "1.0.0" name: paste description: > Paste the latest image or text from the user's local laptop clipboard into this remote Grok session using PasteLocal (SSH tunnel or relay). allowed-tools: Bash(pastelocal-remote:), Read(), Bash(rm:*) metadata: short-description: "Bring laptop screenshots/text into remote Grok via PasteLocal" category: "clipboard"


PasteLocal /paste — Grok Skill

Bring your laptop's clipboard (especially screenshots) into a remote machine where Grok is running.

When to use

  • User says "paste the screenshot", "grab what's on my clipboard", "show me the image from my laptop", etc.
  • Works over existing SSH (preferred) or the experimental relay.

Steps (execute in order)

  1. Try the SSH-tunnel path first (fastest, most common):
  • Run pastelocal-remote via the Bash tool.
  • It will print a file path on success (e.g. /home/user/.cache/pastelocal/pastelocal-abc123.png).
  1. If the SSH path fails or user explicitly wants relay (no direct tunnel or multi-device scenario):
  • Run pastelocal-remote --relay <url> (user must provide the relay URL or it is in their config).
  • Or pastelocal-remote --relay http://localhost:7332 --list to see options, then fetch a specific one.
  • The command still prints a file path.
  1. On any failure (non-zero exit): Report the exact stderr to the user verbatim and stop. Do not retry or guess.
  1. Read the file using your Read tool on the exact path returned.
  1. VisionPaste enrichment (v2 proactive & cached): If the clipboard item was an image/screenshot and the local daemon has [vision] analysis configured (e.g. tesseract OCR), a companion sidecar file with the same base name but .analysis.txt suffix will exist next to the image (e.g. ...-abc123.analysis.txt). Read it too — it contains pre-extracted OCR text and/or a natural-language description (now produced proactively by watcher for instant availability). Present this rich text context to the model first (before or alongside the image); it dramatically improves token efficiency and accuracy for code, errors, UI, diagrams etc. The image remains available via your vision capabilities for anything the text missed.
  1. Clean up: Immediately run rm <path> and (if present) rm <analysis-sidecar> (use Bash) so temp files do not linger.
  1. Confirm with one short line:
  • Image: "Got it, image attached from your laptop clipboard."
  • Text: "Got it, text attached from your laptop clipboard."
  • If it came via relay: "Got it via relay, image attached."

Tips for great UX

  • If the user has multiple devices paired via relay, you may need to ask "which one?" and use --list.
  • With VisionPaste enabled on the user's daemon, screenshots arrive with rich OCR + description context already attached — use it immediately instead of spending extra turns on vision analysis.
  • For repeated use, suggest the user saves a named snippet locally with pastelocal snippets save my-design.

This skill is the Grok-native equivalent of the Claude ~/.claude/commands/paste.md. It prefers the zero-config SSH experience when available.

All versions