The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines methods for processing untrusted external data, which creates a vulnerability surface if the agent implements these patterns without strict validation.
Ingestion points: The documentation includes entry points for untrusted data via [SpacetimeDB.Reducer], [SpacetimeDB.Procedure], and [SpacetimeDB.HttpHandler] arguments (e.g., string name, string input, and HttpRequest request) in SKILL.md.
Boundary markers: There are no instructions or examples provided for using delimiters or explicit 'ignore embedded instructions' warnings for external data.
Capability inventory: The SDK reference describes capabilities for database writes (ctx.Db.Insert, Update, Delete) and outbound network requests (ctx.Http.Get, ctx.Http.Send).
Sanitization: The provided C# snippets do not include input validation, sanitization, or schema enforcement for the incoming data strings or payloads.
[DATA_EXFILTRATION]: The SDK documentation describes the ctx.Http API, which enables the creation of outbound network requests (Get, Send). While this is a standard feature of the developer SDK, it provides the functional capability for data to be transmitted to external URLs.

csharp-server