The Agent Skills Directory

[SAFE]: The skill provides defensive logic and templates for AI safety. It includes regex-based detection for prompt injection and PII, schema validation using Pydantic, and hallucination checks via LLM-based grounding verification. All operations are local string processing or calls to well-known AI services.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing untrusted user input and LLM responses to perform safety checks.
Ingestion points: The check_patterns, check_semantic, and check_pii functions in SKILL.md accept external text inputs.
Boundary markers: The skill acts as the boundary itself; no additional delimiters are present in the code snippets.
Capability inventory: Capabilities are limited to regex matching, string manipulation, and standard LLM API requests. No dangerous system calls or file operations are present.
Sanitization: The skill provides sanitization through its scrub_pii and InjectionDetector logic.

ai-guardrails