The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions define how the agent should interpret and respond to feedback signals, creating a surface for indirect prompt injection from external inputs.
Ingestion points: The agent is instructed to process data from the design_feedback tool, which includes human-provided chat messages and element annotations (referenced in SKILL.md and references/_conversation.md).
Boundary markers: The provided rules do not include specific delimiters or instructions to ignore potential injection patterns within the ingested feedback.
Capability inventory: The agent utilizes tools such as design_create, diagram_upsert, design_feedback, and design_export to modify diagrams and export designs based on input.
Sanitization: The skill lacks explicit guidance for sanitizing or validating external feedback before it is used to drive subsequent tool calls.

diagram-quality