The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions specify that the "Agent Prompt Guide" section in the external DESIGN.md file overrides the skill's own instructions. This design creates a direct pathway for indirect prompt injection if the design file originates from an untrusted source.
Ingestion points: The agent is instructed to read DESIGN.md from the current working directory before starting the workflow.
Boundary markers: The instructions lack explicit boundaries, delimiters, or warnings to ignore malicious commands embedded within the external content.
Capability inventory: The skill utilizes the file_write capability to produce the final index.html landing page.
Sanitization: There is no logic provided to sanitize, escape, or validate the content extracted from the external "Agent Prompt Guide" before it influences the agent's behavior.

saas-landing