The Agent Skills Directory

Indirect Prompt Injection Consideration: The skill is designed to process content from external websites via subagents and the fetch_url tool, which could potentially introduce untrusted instructions into the agent's context.
Ingestion points: External data enters the workflow through the fetch_url tool and subagent findings stored in local files (findings_*.md) derived from web_search results.
Boundary markers: The provided instructions do not specify the use of delimiters or specific warnings to differentiate retrieved web content from system instructions.
Capability inventory: The skill utilizes write_file, read_file, list_files, fetch_url, and the task tool for spawning subagents.
Sanitization: No explicit steps for sanitizing, filtering, or validating the external content are included in the synthesis process.
Filesystem Interaction: The skill instructs the agent to manage local research data by creating directories and files.
Evidence: The workflow uses mkdir to create research folders and write_file/read_file to manage research plans and findings within the local working directory.
Subagent Orchestration: The skill uses the task tool to delegate research tasks to separate agent instances.
Evidence: Instructions guide the agent to spawn subagents with specific web_search budgets and instructions to save output to local files.

web-research