The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites during the discovery and scraping phases (Phase 2 and Phase 3) using Tavily and WebFetch.
Ingestion points: External URLs discovered from homepages and manual fallbacks are scraped and their content is stored in knowledge buckets (SKILL.md, Phase 2/3).
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions embedded within the scraped web content.
Capability inventory: The skill has significant capabilities, including network operations via curl to CRM and MCP endpoints, writing to knowledge buckets (add_to_bucket), agent configuration updates (update_widget_style), and execution of local Python scripts.
Sanitization: The skill mentions structural cleaning (removing navigation, footers, and cookie banners) but does not include explicit prompt sanitization or escaping of the scraped text to prevent embedded instructions from influencing the agent.
[DYNAMIC_EXECUTION]: The skill instructs the user to execute a local Python script (scripts/refurbish-all.py) for processing large batches of demos (Batch Script section). While the script is local, runtime execution of external scripts poses a risk if the script content is modified or if it handles input insecurely.
[COMMAND_EXECUTION]: The skill relies heavily on shell commands (curl) to interact with the vendor's infrastructure at psquared.dev. It also provides specific SQL commands for manual execution to clean up orphaned database records in Supabase (Known Issues & Workarounds section).

refurbish-demos