The Agent Skills Directory

[Privilege Escalation] (HIGH): The skill requests Bash and Task in its allowed-tools list. Granting an agent command execution capabilities for a simple collection of schema patterns is a violation of the principle of least privilege and introduces significant security risk.
[Metadata Poisoning] (MEDIUM): The skill claims to document 'Zod 4', which is currently fictional or non-standard, and provides 'NEW' syntax (e.g., z.email()) that contradicts the standard library. This deceptive metadata can cause the agent to generate broken code and behave unpredictably.
[Indirect Prompt Injection] (HIGH): Mandatory Evidence Chain:
Ingestion points: Zod schemas are primarily used to parse and validate untrusted external data from users or APIs.
Boundary markers: None identified in the skill instructions.
Capability inventory: The skill requests Bash, Write, WebFetch, and Task permissions.
Sanitization: None provided for the data being parsed.
Analysis: By combining the processing of untrusted data with high-privilege execution and network tools, the skill creates a dangerous surface for indirect prompt injection attacks.
[Unverifiable Dependencies & Remote Code Execution] (MEDIUM): The inclusion of WebFetch and WebSearch in allowed-tools for a pattern library allows the agent to ingest external content that could contain further malicious instructions or lead to data exfiltration.

zod-4