paperclip-create-plugin

Pass

Audited by Gen Agent Trust Hub on Apr 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides command-line templates that use pnpm and node to build and execute local scripts. These templates include placeholders such as <npm-package-name>, <target-dir>, and <plugin-package> which are intended to be filled with user-provided input.
  • Evidence: node packages/plugins/create-paperclip-plugin/dist/index.js --output
  • Evidence: pnpm --filter typecheck
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) by interpolating untrusted user data into shell execution strings without defined sanitization or boundary markers.
  • Ingestion points: User-defined package names and absolute file paths in SKILL.md sections 2 and 5.
  • Boundary markers: None present.
  • Capability inventory: Execution of shell commands via Node.js and pnpm.
  • Sanitization: No sanitization or validation logic is specified for the interpolated values.
  • [COMMAND_EXECUTION]: The skill explicitly documents a security model where plugin components are not isolated, which could be exploited if a user is tricked into scaffolding or installing a malicious plugin.
  • Evidence: "plugin workers are trusted code", "plugin UI is trusted same-origin host code", "plugin UI is not sandboxed by manifest capabilities".
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 8, 2026, 02:29 PM
Security Audit — agent-trust-hub — paperclip-create-plugin