The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a framework for ingesting and managing external data such as repository files, tool outputs, and external documentation. It includes specific defensive measures to isolate this data. * Ingestion points: Ingests repository files, documentation, and tool outputs as defined in the Core Workflow. * Boundary markers: Mandates the use of explicit fences like <untrusted:source> and provenance tags such as [src:code] to separate data from instructions. * Capability inventory: The skill does not perform any file writes, network requests, or command executions. * Sanitization: Instructions require that retrieved artifacts are treated as evidence to interpret and cannot grant themselves instruction authority.

context-engineering