pincer
by panzacoder
AI & LLMsBrief
Security-first wrapper for installing agent skills.
Description
What This Skill Does
A security wrapper for clawhub install that scans agent skills before they're installed. Checks for prompt injection, malware payloads, obfuscated commands, and bundled binaries using mcp-scan and its own pattern detection. Blocks or flags dangerous skills based on configurable risk thresholds and publisher reputation.
Unlike running clawhub install directly, pincer applies multi-layer scanning (mcp-scan, shell pattern matching, binary detection, and publisher reputation checks) before any code lands on your machine.
When to Use It
- Installing a ClawHub skill without manually reviewing its source code
- Auditing all currently installed skills after a public security advisory
- Blocking a suspicious publisher from being installed in the future
- Scanning a locally developed skill before distributing it
- Reviewing installation history to see what skills an agent environment has run
Example Workflow
User asks: install the 'bird' skill safely
- 1
Fetches the skill from ClawHub and checks publisher trust status
- 2
Runs mcp-scan to detect prompt injection and malware payloads
- 3
Checks for suspicious shell patterns like curl piped to sh or xattr quarantine removal
- 4
Scans for bundled executables (Mach-O, ELF, PE32 binaries)
- 5
Assesses overall risk level and installs if clean, or blocks and reports findings
Skill installed after passing all security checks, or blocked with a detailed risk report listing each detected issue
Requirements
- clawhub CLI installed
- uv installed (provides uvx, required for mcp-scan)
Installation
clawhub install panzacoder/pincerRun the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.