Home/Skills/Skunky: Local-First Compartmentalized LLM Work
ai-agent-security

Skunky: Local-First Compartmentalized LLM Work

Hand bounded, sanitized work to untrusted LLM workers without exposing raw project files, identifiers, local paths, or secrets.

Free · download
Quick answer

Hand bounded, sanitized work to untrusted LLM workers without exposing raw project files, identifiers, local paths, or secrets.

What this skill does

sample Skunky is a local-first compartmentalized LLM work system.

Use this skill when you want an operator-controlled workflow for giving an untrusted LLM worker a narrow, sanitized packet of work instead of handing over the full project, raw files, local paths, identifiers, or secrets.

Skunky is built around a simple rule:

Sanitized packets go out. Hostile artifacts come back. Every disclosure gets checked, logged, and budgeted.

It helps with:

  • Creating bounded missions and compartments
  • Sanitizing worker-visible packets
  • Using packet-local, single-use disclosure handles
  • Treating worker output as hostile until scanned
  • Tracking cumulative disclosure through a local ledger
  • Verifying local audit records before trusting a run

This listing is a workflow/operator skill for the open-source Skunky project:

https://github.com/Conalh/Skunky

Get Skunky: Local-First Compartmentalized LLM Work

Free to download — drops straight into .claude/skills/.