The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill configuration uses npx -y @pactortester/figma-mcp-cached to download and execute code from a third-party source at runtime. The package belongs to a user ('pactortester') that does not match the stated author ('Aradotso') and is not from a known trusted organization.
[COMMAND_EXECUTION]: The MCP server is initialized via a shell command (npx), which executes the downloaded package with full access to the agent's execution environment permissions.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill handles sensitive Figma Personal Access Tokens (FIGMA_API_KEY) and writes design data to local cache directories (e.g., ~/Library/Caches/FigmaMcp). While these are necessary for the skill's function, they represent a surface for potential data exposure if the cache is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted design data from Figma URLs, creating a surface for injection.
Ingestion points: Figma API responses (design nodes, text, and styles) ingested via get_figma_data in SKILL.md.
Boundary markers: None identified; instructions do not specify the use of delimiters or warnings to ignore instructions within design data.
Capability inventory: Subprocess execution (npx), file-system writes (download_figma_images to ~/Downloads and disk caching), and network access to the Figma API.
Sanitization: No evidence of sanitization, validation, or escaping of design content before it is provided to the agent context.

figma-mcp-cached