The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill reads user-supplied error logs and descriptions which can be manipulated to trigger malicious actions in an agent equipped with shell execution capabilities.
Ingestion points: User input regarding failures and log snippets (SKILL.md).
Boundary markers: Absent. No markers or safety instructions are provided to isolate untrusted user data.
Capability inventory: Subprocess calls for npm, npx, mkdir, and chmod (SKILL.md, ERROR-REFERENCE.md).
Sanitization: Absent.
[Remote Code Execution] (HIGH): The skill uses npx to download and execute code from unverified packages. Evidence: npx zodipus inspect and npx zodipus generate (SKILL.md).
[External Downloads] (MEDIUM): The skill installs zodipus and prisma packages from non-whitelisted sources. Evidence: npm install zodipus and npm update prisma (SKILL.md, ERROR-REFERENCE.md).
[Command Execution] (MEDIUM): The skill performs file system modifications and permission updates on local directories. Evidence: mkdir -p ./generated and chmod 755 ./generated (ERROR-REFERENCE.md).

zodipus-troubleshooting