The Agent Skills Directory

[Dynamic Execution] (MEDIUM): The search workflow in SKILL.md uses a shell command that executes a python snippet: $(python3 -c "import urllib.parse; print(urllib.parse.quote('YOUR QUERY'))"). There is a risk of command injection if the agent interpolates user-provided search terms containing single quotes without proper escaping, which could allow arbitrary code execution.
[Indirect Prompt Injection] (LOW): The skill ingests documentation content from developerknowledge.googleapis.com. This creates an indirect prompt injection surface. Evidence from SKILL.md: Ingestion points are the API responses; boundary markers and sanitization are absent; capabilities include command execution and network access.
[Data Exposure & Exfiltration] (SAFE): The skill performs network requests to an official Google API. This is a trusted source and is considered safe as it does not access or exfiltrate sensitive local data.

google-devdocs