The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest data from external websites using tools like curl and playwright. This creates a surface for indirect prompt injection as untrusted content from the web is brought into the agent's context.
Ingestion points: The skill body in SKILL.md provides code snippets to fetch raw HTML and API responses.
Boundary markers: The provided instructions do not include explicit delimiters or warnings to treat scraped content as untrusted.
Capability inventory: The skill utilizes network access (curl) and automated browser execution (playwright).
Sanitization: No sanitization or validation of the external content is implemented in the provided examples.
[COMMAND_EXECUTION]: The documentation provides multiple shell command examples, including curl for fetching web content and grep for parsing JSON blobs or HTML markers.
[EXTERNAL_DOWNLOADS]: The skill contains instructions to install a Model Context Protocol (MCP) server using the command claude mcp add --transport http manypi https://mcp.manypi.com/mcp -s user. This targets the vendor's own infrastructure for expanding agent capabilities.

javascript-rendered-scraping