The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a workflow that is susceptible to indirect prompt injection. It explicitly instructs the agent to read an external file (DESIGN.md) and grants any instructions found within a specific 'Agent Prompt Guide' section the power to override the skill's core behavior.
Ingestion points: The skill reads DESIGN.md from the current working directory as the first step of its execution workflow.
Boundary markers: The instructions lack delimiters or specific directives to treat the ingested design content as non-executable data, potentially allowing embedded instructions to be obeyed by the agent.
Capability inventory: The skill uses the file_write capability to create the final index.html file.
Sanitization: There is no evidence of validation or sanitization of the content provided in the design file before it is processed by the agent's logic.

saas-landing