The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an interface to browse and scrape arbitrary websites via unbrowse.scrape and unbrowse.browse.snapshot, creating a surface for processing untrusted content.
Ingestion points: External data enters the agent context through the scraping and snapshot tools defined in references/tools.json.
Boundary markers: The SKILL.md file contains explicit behavioral constraints, instructing the agent to report completion only from verified results and to obtain authorization for any external writes.
Capability inventory: The skill allows for complex web interactions including form filling and clicking through unbrowse.browse.act and unbrowse.run.
Sanitization: The instructions focus on structured data handling and human review rather than specific content filtering.
[DYNAMIC_EXECUTION]: The unbrowse.learn tool dynamically generates new capabilities and agent skill definitions based on browser traces or HAR recordings, which are then utilized by the unbrowse.run executor.
[COMMAND_EXECUTION]: The skill documentation provides standard setup instructions using CLI commands for MCP configuration (claude mcp add) and local management via the unbrowse CLI.

unbrowse