The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to interact with and extract data from external, untrusted web pages, creating a surface for indirect prompt injection attacks where malicious content on a website could influence agent behavior.
Ingestion points: The skill ingests untrusted data from URLs processed via firecrawl scrape (SKILL.md).
Boundary markers: There are no explicit instructions or delimiters mentioned to isolate web content from agent instructions.
Capability inventory: The skill uses Bash(firecrawl *) and Bash(npx firecrawl-cli *) to click elements, fill forms, and execute browser-based code (SKILL.md).
Sanitization: No sanitization or validation mechanisms are described for the scraped content before it is processed by the interaction layer.
[COMMAND_EXECUTION]: The skill relies on executing shell commands through firecrawl and npx firecrawl-cli to perform its primary functions.
[EXTERNAL_DOWNLOADS]: The skill references a vendor-owned GitHub repository (github.com/firecrawl/skills) and utilizes npx to execute the firecrawl-cli package, which involves fetching code from the npm registry.
[DYNAMIC_EXECUTION]: The skill supports an --code flag that allows the agent to provide and execute arbitrary browser-automation code (e.g., agent-browser click @e5) during the interaction session.

firecrawl-interact