The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the web through multiple automated agent phases, creating a significant injection surface.
Ingestion points: scripts/workflow-script.js uses WebFetch inside the FETCH_PROMPT to retrieve content from external URLs provided by search results.
Boundary markers: The prompts (e.g., FETCH_PROMPT, VERIFY_PROMPT) interpolate raw source content or quotes without using clear delimiters (like XML tags or triple quotes) or 'ignore' instructions for the processing agents.
Capability inventory: The skill uses WebSearch and WebFetch tools to interact with the network. While it doesn't directly write files or execute local shell commands, its output influences the agent's research report and final response to the user.
Sanitization: There is no evidence of HTML stripping, sanitization, or filtering of the fetched content before it is passed to the next agent in the pipeline.
[COMMAND_EXECUTION]: The skill makes extensive use of the agent() function to execute complex, multi-stage logic using structured outputs and tool calls (WebSearch, WebFetch).
[EXTERNAL_DOWNLOADS]: The skill dynamically identifies and fetches content from a large number of external websites (up to 15 sources across 5 search angles) to synthesize information.

deep-research