Zum Inhalt springen
TakoAPI
Zurück zu den Skills

pipelock

von luckypipewrench

Browser & Automation
—Noch nicht bewertet
Deine Bewertung

Kurzbeschreibung

Secure agent HTTP requests through a scanning proxy that catches credential leaks, SSRF, and prompt injection.

Beschreibung

Was dieser Skill macht

Pipelock is a fetch proxy that intercepts outbound HTTP requests from AI agents and runs them through a 7-layer scanner pipeline. It detects credential leaks, SSRF attempts, prompt injection, data exfiltration, and env var exposure before requests leave the machine. Configure responses as block, strip, warn, or interactive approval.

Running security checks at the network layer catches threats regardless of which agent framework or tool makes the request, without requiring code changes in each tool.

Wann du ihn einsetzt

  • Blocking an agent from leaking API keys in request bodies
  • Preventing SSRF when an agent fetches user-supplied URLs
  • Scanning MCP server responses for prompt injection
  • Auditing all outbound agent traffic without blocking
  • Detecting unauthorized workspace file changes during agent runs

Beispiel-Workflow

EINGABE

User asks: wrap my filesystem MCP server with security scanning

AGENT
  1. 1

    Generate a pipelock config with the balanced preset: pipelock generate config --preset balanced -o pipelock.yaml

  2. 2

    Start pipelock proxy: pipelock run --config pipelock.yaml

  3. 3

    Wrap the MCP server: pipelock mcp proxy -- npx @modelcontextprotocol/server-filesystem /path/to/dir

  4. 4

    Agent requests now pass through the 7-layer scanner before reaching the network

  5. 5

    Any detected API key leak or prompt injection triggers the configured action (block/strip/warn/ask)

AUSGABE

MCP server running behind pipelock proxy with credential leak and prompt injection detection active

Voraussetzungen

  • pipelock binary (via brew install luckyPipewrench/tap/pipelock or go install)

Installation

clawhub install luckypipewrench/pipelock

Führe den CLI-Befehl aus oder füge den Prompt in Claude Code, Cursor, Codex oder deinen Coding-Agenten ein.

Links