The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the vendor's CLI tool (@sitegpt/cli) from the npm registry. It also uses curl to fetch HTML content and image assets from user-provided URLs to the /tmp/ directory for inspection and chatbot configuration.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Fetches and processes data from external, untrusted websites using curl commands (SKILL.md).
Boundary markers: Explicitly instructs the agent to create a discovery brief and wait for human confirmation before creating or training the chatbot.
Capability inventory: The skill can execute shell commands (curl, npm, sitegpt), write configuration files (persona.md, instructions.md), and manage SiteGPT account resources via the CLI.
Sanitization: There is no automated sanitization of the website content; instead, the workflow relies on a human-in-the-loop review step to verify the agent's inferences.
[COMMAND_EXECUTION]: The skill uses shell commands to install software, inspect website headers and metadata, and interact with the SiteGPT service. While automated scans flagged these as potential remote code execution, manual review confirms the downloaded content is used for analysis and asset upload rather than being executed as code.

sitegpt-cli