The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, including Google Places JSON and remote official restaurant websites, to generate copy and CMS content. This creates a surface for indirect prompt injection where malicious content on a restaurant's website could attempt to influence the agent's behavior during the site-building process.\n
Ingestion points: places/${placeId}.json and remote website URLs fetched during the extraction phase.\n
Boundary markers: The skill instructions include a "Customer Visibility Gate" to filter content, but do not define explicit prompt delimiters for the external data.\n
Capability inventory: The agent has access to file system operations, network requests, and shell command execution (curl, npx dineway, sqlite3).\n
Sanitization: Content generation involves paraphrasing source material rather than direct interpolation, providing basic mitigation.\n- [EXTERNAL_DOWNLOADS]: The scripts/restaurant_site_data.js utility downloads restaurant images from remote URLs discovered in the place data. While the script includes a filter (isUsableImageUrl) for common image hosting domains, it performs arbitrary network requests to fetch these assets.\n- [COMMAND_EXECUTION]: The skill utilizes shell commands for validation and CMS management. Specifically, the scripts/restaurant_site_data.js helper script uses spawnSync to execute the dineway CLI for media uploads, and the main workflow uses curl and sqlite3 for site validation.

dineway-building-restaurant