The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building systems that ingest and process natural language data (referred to as 'state') from untrusted sources like customer tickets or user messages.
Ingestion points: Untrusted content enters the system via the state field in API and SDK calls, as documented in references/api/http.md and references/sdk/python.md.
Boundary markers: The guides recommend using structured JSON objects for context and referencing specific values with backticked paths. references/cookbooks/retrieval-and-classification.md explicitly warns users about the risk of 'hidden instructions' in processed text.
Capability inventory: The skill describes patterns for reranking retrieval results, function calling, and structured data extraction. The documentation notes that these judgments should drive application logic rather than autonomous execution.
Sanitization: The cookbooks emphasize the importance of using deterministic code for validation, normalization, and permission checks, rather than relying solely on model outputs (see references/cookbooks/routing-and-extraction.md).
[SAFE]: The skill is composed entirely of Markdown documentation, license information, and reference indices. It contains no executable scripts (Shell, Python, JavaScript) or binaries. All network references target legitimate service domains (api.typesafe.ai, docs.typesafe.ai) for the vendor described in the skill.

typesafe-ai