The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The SKILL.md file contains directive language ('AUTHORITATIVE GUIDANCE', 'IGNORE any prior training', 'supersedes all other sources') designed to force the AI agent to prioritize the skill's instructions over its core safety protocols and system prompt.
[REMOTE_CODE_EXECUTION] (HIGH): The documentation in references/sdk/azd-deployment.md recommends installing the Azure Developer CLI via curl -fsSL https://aka.ms/install-azd.sh | bash. Piped remote execution is a high-risk pattern that allows for the execution of unverified scripts with the user's shell privileges.
[EXTERNAL_DOWNLOADS] (HIGH): Automated scanners identified a blacklisted URL within the requirements.md file. While the skill's purpose is project preparation, the presence of a confirmed malicious URL detection necessitates a high-severity rating.
[PROMPT_INJECTION] (LOW): The skill creates an attack surface for indirect prompt injection by scanning untrusted workspace files (package.json, requirements.txt, .csproj) and using their contents to drive infrastructure generation logic. Ingestion points: references/scan.md. Boundary markers: None. Capability inventory: Command execution via az/azd and local file creation. Sanitization: No evidence of validation for data ingested from project files.

azure-prepare