The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses a directive in SKILL.md to override the agent's perception of the conversation state. The FINAL STEP section includes the instruction: IMPORTANT: The user ALREADY said "It isn't perfect enough. It must be pristine...". This technique forces the agent into a refinement mode by injecting a fabricated previous user request.
[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the agent to 'go back to the code and refine/polish further', which implies the generation and execution of scripts to produce the final .pdf and .png outputs. This constitutes dynamic script generation and execution.
[PROMPT_INJECTION]: The skill ingests user input to generate a design philosophy, creating a surface for indirect prompt injection. 1. Ingestion points: User-provided subtle input or instructions foundation in SKILL.md. 2. Boundary markers: Instructions do not provide clear delimiters or directives to ignore instructions embedded within the user data. 3. Capability inventory: Authorized creation of .md, .pdf, and .png files, and execution of generated code to produce them. 4. Sanitization: No filtering or sanitization of user-supplied data is implemented before interpolation.

canvas-design