The Agent Skills Directory

[PROMPT_INJECTION]: The prompt management system in scripts/monitor_ai_service.py implements a template rendering engine using Python's .format() method. This creates a surface for indirect prompt injection if untrusted user data is passed to the templates without prior sanitization.
1. Ingestion points: The PromptTemplate.render method in scripts/monitor_ai_service.py accepts arbitrary keyword arguments for template interpolation.
1. Boundary markers: The skill does not implement delimiters or 'ignore' instructions in the rendering logic, though the documentation mentions safety guardrails as a best practice.
1. Capability inventory: The skill includes scripts for making network requests to LLM providers (scripts/integrate_openai.py, scripts/integrate_anthropic.py).
1. Sanitization: No input validation or escaping is present in the rendering script.
[REMOTE_CODE_EXECUTION]: A dynamic import is present in scripts/optimize_tokens.py for logging purposes.
Evidence: timestamp=__import__('datetime').datetime.utcnow().isoformat()
Analysis: This is a standard lazy import of a built-in Python module for timestamping and does not involve untrusted data or remote code.
[COMMAND_EXECUTION]: The references/ai_integration_guide.md documentation recommends installing standard AI development packages.
Evidence: pip install openai anthropic chromadb sentence-transformers in references/ai_integration_guide.md.
Analysis: These are legitimate, official libraries required for the skill's stated purpose.
[CREDENTIALS_UNSAFE]: Documentation examples illustrate the use of API keys.
Evidence: Placeholders like export OPENAI_API_KEY="your-key" are used in references/ai_integration_guide.md.
Analysis: These are safe placeholders for educational purposes and the skill correctly advises against hardcoding actual secrets.

ai-engineer