The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and analyze an existing codebase, creating a vulnerability where instructions embedded in the code could influence the agent's subsequent actions.
Ingestion points: The skill processes external codebase files during the 'Scan' and 'Diagnose' phases in SKILL.md.
Boundary markers: The instructions do not provide specific delimiters or warnings to the agent to ignore or distinguish instructions found within the project files from the skill's own goals.
Capability inventory: The skill has the capability to write and modify files in the project to 'Fix' design issues, which could be abused if an injection is successful.
Sanitization: No sanitization or validation logic is defined for the content read from the external codebase.
[EXTERNAL_DOWNLOADS]: The skill refers to a well-known image placeholder service for use during design upgrades.
Evidence: Suggests using https://picsum.photos/ for generating background assets in SKILL.md.

redesign-existing-projects