The Agent Skills Directory

[PROMPT_INJECTION]: The FINAL STEP section in SKILL.md contains instructions that simulate prior user feedback ("The user ALREADY said 'It isn't perfect enough...' ") to override the current agent context and enforce a specific refinement loop.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user input to generate design philosophies and artifacts, creating an exposure to potentially malicious instructions in the processed data.
Ingestion points: User request and reference deduction in SKILL.md.
Boundary markers: None present; instructions favor creative freedom and subtle reference deduction.
Capability inventory: File creation (PDF, PNG, MD) and local directory searching.
Sanitization: Absent.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: Instructions in SKILL.md suggest the agent "Download and use whatever fonts are needed," promoting external network requests.

canvas-design