The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting data from untrusted local project files to compose model instructions.
Ingestion points: Project files such as brand-guideline.md, prd.md, 02-campaign-brief-global/output.md, package.json, and README.md are read and parsed for brand identity and style context.
Capability inventory: The agent can perform network requests to the OpenAI API and execute shell commands for environment detection and graphic rendering.
Boundary markers: The skill uses a structured 5-part prompt framework, but lacks explicit boundary instructions to ignore potentially malicious commands embedded in the ingested project files.
Sanitization: The skill performs basic validation (such as regex for hex codes) but does not implement comprehensive sanitization of text extracted from files before it is interpolated into model prompts.
[COMMAND_EXECUTION]: The skill utilizes shell script snippets for runtime environment detection.
Evidence: In SKILL.md and CONVENTIONS.md, Bash logic is used to check for the presence of the OPENAI_API_KEY and the OD_BIN executable to determine the operational tier (Free, Pro, or Enterprise).
[REMOTE_CODE_EXECUTION]: The skill references the use of npx for executing remote packages.
Evidence: The references/quote-graphic.md and examples/ai-income-mastery-quote-graphic.md files describe a workflow using npx puppeteer-screenshot to render HTML templates into PNG assets.
[EXTERNAL_DOWNLOADS]: The skill references resources from well-known and trusted external services.
Evidence: The HTML templates for quote graphics (templates/quote-graphic.md) fetch typography assets from Google's official font service (fonts.googleapis.com).
Evidence: The skill provides links to official Canva templates for users needing text-heavy infographics.

30-design-master-global