The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent on creating web handlers that ingest and process untrusted external data from HTTP requests, which is a standard attack surface for web applications.
Ingestion points: Untrusted data enters the agent context via Gin's binding methods like ShouldBindJSON, ShouldBind, ShouldBindUri, and ShouldBindPlain, as well as path parameters using c.Param, as described across SKILL.md and the version-specific reference files.
Boundary markers: The instructions emphasize using ShouldBind* methods which leverage binding tags for validation, providing a level of structure to ingested data, though it does not explicitly define markers for preventing LLM-specific injection when this data is used in prompts.
Capability inventory: Handlers developed following these instructions are expected to interact with downstream systems such as databases and HTTP clients (referenced in the 'Context' section of SKILL.md) and may perform file operations (referenced in the 'Static Files' section of references/1.11.md).
Sanitization: The skill promotes sanitization through the use of the go-playground/validator library via struct tags mentioned in the 'Binding and Response' section of SKILL.md.

gin