The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill defines a logic-driven process for ingesting external data to generate agent prompts.
Ingestion points: Uses read() operations on templates/subagent-template.md, reference/patterns.md, and reference/integration-patterns.md to construct a draft prompt.
Boundary markers: No specific delimiters or boundary instructions are defined in the symbolic logic to prevent instructions within those files from overriding the agent's behavior.
Capability inventory: The skill references a scripts/ directory for 'validation and metrics automation' and suggests the use of 'MCP tools,' which are not included in the provided file but represent the skill's operational surface.
Sanitization: The logic focuses on metrics like 'integration_score' and 'clarity' but does not specify any sanitization or escaping of the ingested text content.

subagent-prompt-construction