The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a command-line tool named getnote to save, retrieve, and manage notes. This usage is consistent with the skill's documented purpose for content collection.
[DATA_EXFILTRATION]: The instructions include a specific security constraint: 'Keep browser discovery read-only. Do not inspect cookies, local storage, profile files, passwords, or session stores.' This proactively prevents the agent from harvesting sensitive user credentials during the browser automation phase.
[INDIRECT_PROMPT_INJECTION]: As the skill ingests content from external web pages (Xiaohongshu note titles and transcripts), there is a standard risk that maliciously crafted content on the platform could influence the agent's summary.
Ingestion points: Data enters the context via the collectVisibleCards snippet in references/chrome-collector.md and the getnote note command.
Boundary markers: The instructions do not define strict delimiters for the external content but emphasize using web_content for analysis rather than the tool's generated summaries.
Capability inventory: The skill can execute getnote CLI commands and browser automation scripts via a Chrome plugin.
Sanitization: The skill instructs the agent to clean transcript errors and synthesize findings rather than verbatim copying.

xhs-note-batch-analysis