The Agent Skills Directory

[SAFE]: The skill consists of documentation and system prompt templates that adhere to security best practices.
[SAFE]: It explicitly warns against including sensitive data in external queries and advocates for minimal tool permissions.
[INDIRECT_PROMPT_INJECTION]: The skill describes workflows where agents process external project data. It provides mitigations by teaching users to implement XML boundary markers, explicit constraints, and output validation to prevent the execution of malicious instructions found in data. 1. Ingestion points: Agent templates consume project files and documentation via the Read tool. 2. Boundary markers: The skill mandates the use of semantic XML tags to structure prompts and separate instructions from data. 3. Capability inventory: Templates include access to Read, Write, Edit, and Bash tools depending on the agent's role. 4. Sanitization: Recommends using specialized validator agents and implementing strict output schema checks.

agents-manager