The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary functionality is delivered through the execution of the agbrowse CLI tool and its subcommands (search, fetch, extract, research). This includes capabilities for browser automation and interaction.
[DATA_EXPOSURE]: The skill implements an 'Intent Guard' that strictly separates web search tasks from local repository tasks. It explicitly instructs the agent to use local tools (like rg) for repository code or logs and forbids routing such data to web search tools.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a natural attack surface for indirect prompt injection by ingesting and processing untrusted web content.
Ingestion points: Web content fetched via agbrowse search and agbrowse fetch (references/evidence-ladder.md).
Boundary markers: Uses structured JSON envelopes (agbrowse-search-v1) to separate fetched content (textExcerpt) from the agent's logic.
Capability inventory: Execution of agbrowse subcommands for searching, fetching, and extracting data.
Sanitization: Implements specialized extraction (Defuddle), content scoring (content-scorer.mjs), and SSRF/DNS-rebinding guards to mitigate risks from malicious pages.
[SAFE]: The skill documentation details security-positive features in the underlying tool, including the rejection of credential-bearing URLs and the blocking of private network IP addresses.

search