The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to orchestrate a research pipeline. This includes running the browse CLI for web operations, executing local Node.js scripts (extract_page.mjs, compile_report.mjs, list_urls.mjs) for data processing, and using heredocs to write markdown files.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the browse CLI via npm (npm install -g browse). This is an official tool from the vendor (Browserbase) and is necessary for the skill's primary functionality of searching and fetching web content.
[INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests untrusted data from external websites (homepages, sitemaps, and search results).
Ingestion points: extract_page.mjs and browse cloud fetch used within the research subagents.
Boundary markers: The subagent prompt templates in references/workflow.md use labeled headers and delimiters to separate context from user data.
Capability inventory: The agents have access to the Bash tool, which is used to call the vendor's CLI and local scripts.
Sanitization: scripts/extract_page.mjs implements HTML cleaning, removing scripts, styles, and other non-visible elements before passing content to the agent, reducing the risk of malicious payload execution.
[DYNAMIC_EXECUTION]: The scripts/compile_report.mjs script utilizes execSync to invoke the system's open command to display the generated HTML report in a browser. This is a standard convenience feature for local reporting tools.

company-research