The Agent Skills Directory

[SAFE]: The skill consists primarily of structured documentation (pattern cards), templates, and examples for architectural design. It does not contain any prompt injection attempts or instructions to bypass safety guidelines.\n- [COMMAND_EXECUTION]: The skill provides a Python script (scripts/lint_agentic_arch.py) for the agent to use during architecture reviews. Analysis of the script confirms it is a legitimate static analysis tool that uses Python's standard library (pathlib, json, re) to inspect markdown files. It does not use dangerous functions like eval(), exec(), or os.system() on user-controlled input.\n- [SAFE]: The skill correctly identifies security risks such as 'Tool Injection' and 'Least Privilege' in its documentation, providing actionable mitigations (e.g., using XML delimiters to separate instructions from data, and restricting tool access to specific agents).\n- [SAFE]: All external references and mentioned libraries (e.g., Pinecone, LangChain, Vertex AI) are well-known, trusted services in the AI ecosystem.

agentic-infra