The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes a synthetic user persona feedback loop ("The user ALREADY said 'It isn't perfect enough...'") within its internal instructions. While intended to enforce high-quality output and refinement, this technique employs a simulated conversation history to override the agent's default operational parameters.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted user instructions to generate its core output.
Ingestion points: The skill ingests user input to serve as the 'foundation' for creating a Design Philosophy (SKILL.md).
Boundary markers: No explicit delimiters, XML tags, or boundary markers are defined to isolate user-provided content from the system prompt's instructions.
Capability inventory: The agent is authorized to write multiple file types including .md, .pdf, and .png, which could be misused if the generated content is influenced by malicious instructions.
Sanitization: The skill does not implement any validation or sanitization procedures for the ingested user data.
[EXTERNAL_DOWNLOADS]: The skill directs the agent to "Download and use whatever fonts are needed." This instruction encourages the acquisition of external assets from unverified locations. Depending on the agent's tool access, this could lead to the download of malicious files or the use of assets from untrusted third-party domains.

canvas-design