michaelschecht OpenClaw Action
Overview
Set up CI security scanning for agent workspaces and skill repositories. Use this when a project needs pull request checks for exposed secrets, prompt injection markers, suspicious shell usage, or data exfiltration patterns in agent files.
What to Scan
skills/**,.claude/**,.codex/**,.openclaw/**,.agents/**, and other agent workspace folders.- Scripts, hooks, install files, and generated instructions bundled with skills.
- Workflow files that grant network, token, or write permissions.
GitHub Actions Pattern
name: Agent Security Scan
on:
pull_request:
paths:
- 'skills/**'
- '.claude/**'
- '.codex/**'
- '.openclaw/**'
push:
branches: [main]
jobs:
scan:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Scan agent workspace
run: python scripts/scan-agent-workspace.py .
Findings to Flag
- API keys, private keys, tokens, passwords, and wallet material.
- Instructions that try to bypass higher-priority rules or reveal secrets.
- Shell snippets that download and execute remote code without verification.
- Unexpected network calls, webhooks, or upload commands.
- Broad file deletion, credential harvesting, or persistence mechanisms.
Operating Principles
The scan should detect and annotate risks without modifying repository files. Fail CI for high-confidence critical findings and emit a readable summary for lower-confidence warnings.
Source Notes
Adapted from the SkillsMP openclaw-action skill, selected because it is a highly starred security automation skill focused on agent workspace supply-chain risk.