🛡️ Security: Add automated security scanning for plugin submissions

[elliotllliu]

🛡️ Security: Add automated security scanning for plugin submissions

Problem

The current plugin pre-check pipeline (pre-check-plugin.yaml) validates manifest format, icons, versioning, dependencies, and packaging — but does not include any security analysis of the plugin code itself.

After scanning all 493 plugins in this repository with AgentShield, we found:

Category	Count	%
🔴 Plugins with High Risk findings	6	1.2%
🟡 Plugins with Medium Risk findings	72	14.6%
🟢 Clean plugins	415	84.2%

High Risk Examples

LogicOber/better-e2b-sandbox (Score: 0/100)

• tools/create-nextjs-bun-sandbox.py:134 — downloads and executes remote code via pipe-to-shell
• tools/install-packages.py:18 — same pattern
• Reverse shell patterns detected in multiple files

allenyzx/enhancing_function_agent (Score: 0/100)

• 4 instances of eval() with dynamic (user-controlled) input in strategies/enhancing_function_agent.py

bowenliang123/md_exporter (Score: 0/100)

• 6 instances of exec() with dynamic input in bundled md2pptx library

The full report with all findings is available here: Dify Plugins Security Report

Suggestion

Add a security scanning step to the pre-check pipeline. For example, using AgentShield:

- name: Security Scan
  run: |
    npx @elliotllliu/agentshield@latest scan "$PLUGIN_PATH" --json > security-report.json
    SCORE=$(node -e "console.log(JSON.parse(require('fs').readFileSync('security-report.json','utf8')).score)")
    HIGH=$(node -e "console.log(JSON.parse(require('fs').readFileSync('security-report.json','utf8')).findings.filter(f=>f.severity==='high').length)")
    echo "Security Score: $SCORE/100, High Risk: $HIGH"
    if [ "$HIGH" -gt 0 ]; then
      echo "::warning::Plugin has $HIGH high-risk security findings (score: $SCORE/100)"
    fi

This would:

• ✅ Zero install required (npx)
• ✅ Scan Python code for eval()/exec(), data exfiltration, backdoors, reverse shells
• ✅ Detect prompt injection in tool descriptions
• ✅ Support .difypkg zip extraction
• ✅ Complete in <1s per plugin
• ✅ Non-blocking (warning only, not failing the build)

Notes

• AgentShield is open source (MIT license): https://github.com/elliotllliu/agentshield
• The scan used v0.3.0 with three-tier risk classification (High/Medium/Low)
• We focused on minimizing false positives — the 6 high-risk findings above are all confirmed real patterns
• Happy to help integrate or adjust rules for the Dify plugin ecosystem