The Agent Skills Directory

[SAFE]: The skill's content is limited to documentation and code templates for building resilient infrastructure. There are no executable scripts, remote downloads, or hidden instructions.
[INDIRECT_PROMPT_INJECTION]: The implementation examples demonstrate a workflow that ingests and processes user-supplied prompts, which is a potential surface for indirect injection attacks.
Ingestion points: The complete method in SKILL.md accepts CompletionParams containing user-controlled content.
Boundary markers: The provided examples do not include delimiters or instructions to ignore embedded commands.
Capability inventory: The logic enables interaction with external LLM services through provider-specific clients.
Sanitization: No input validation or prompt escaping is implemented in the provided architectural examples.

llm-fallback-chains