The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute local utility scripts including recon.mjs, extract_event.mjs, extract_page.mjs, and compile_report.mjs. It also executes the browse CLI tool for web interaction and browser automation. These executions are part of the core functionality for data collection and report generation.
[EXTERNAL_DOWNLOADS]: The skill requires the global installation of the browse CLI (npm install -g browse), which is a legitimate utility provided by the skill's author (browserbase) to facilitate browser-based data extraction.
[DATA_EXPOSURE]: The skill utilizes the BROWSERBASE_API_KEY environment variable for cloud-based browser sessions. Extracted data and research reports are stored locally in the user's Desktop directory within a folder named after the event. No sensitive local configuration files or credentials are accessed.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the web during prospecting. It implements multiple layers of protection to mitigate the surface area of indirect prompt injection:
Ingestion points: External content is fetched in extract_page.mjs and processed by various scripts.
Boundary markers: Subagent prompts defined in references/workflow.md use clear structural delimiters such as CONTEXT:, TOOL RULES:, and ANTI-HALLUCINATION RULES: to separate instructions from data.
Capability inventory: The skill's capabilities are scoped to Bash execution for internal scripts and the browse tool, with filesystem writes limited to a designated output directory.
Sanitization: The compile_report.mjs script performs rigorous HTML and JavaScript escaping (escapeHtml, escapeAttr, escapeJsInAttr) before rendering external content into the final HTML reports.

event-prospecting