The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a pattern where an agent retrieves and evaluates content from a codebase to refine its search criteria. This creates a surface for indirect prompt injection if the files being retrieved contain malicious instructions designed to manipulate the agent's evaluation or retrieval logic.\n- Ingestion points: The retrieveFiles and evaluateRelevance logic in SKILL.md processes content from project files.\n- Boundary markers: No specific delimiters or instructions to ignore embedded commands are included in the provided pattern.\n- Capability inventory: The pattern assumes the agent has capabilities for file searching and reading.\n- Sanitization: The pseudo-code does not demonstrate sanitization or validation of the retrieved file content before it is used to influence the next iteration of the retrieval loop.

iterative-retrieval