The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by scanning existing codebases and possesses the capability to modify files, creating an attack surface for indirect prompt injection.
Ingestion points: The skill instructions in SKILL.md require the agent to read the codebase to identify frameworks, styling methods, and design patterns.
Boundary markers: No explicit delimiters or instructions to ignore embedded prompts in the scanned code are provided.
Capability inventory: The skill allows the agent to "Apply targeted upgrades" to the project files, which necessitates file-write operations.
Sanitization: No sanitization or validation of the content read from the codebase is mentioned.
[EXTERNAL_DOWNLOADS]: The skill references an external URL for assets.
Evidence: The design audit suggests using https://picsum.photos/seed/{name}/1920/1080 for placeholder imagery in SKILL.md. This is a well-known service for developers and is considered a safe reference.

redesign-existing-projects