integrate-napkin-agent-skills

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). SKILL.md’s required runtime workflow reads outsider-authored free text by having the agent inspect and then retrieve content from a local Napkin vault (e.g., reading NAPKIN.md and running napkin overview/search/read) which can include arbitrary user-submitted markdown in that vault.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill includes explicit runtime installation commands that fetch and execute external packages which can inject context and background writes into agent sessions (e.g., "pi install npm:pi-napkin -l --approve" and "npm install -g napkin-ai"), so external code would run and could directly influence prompts/behavior.