The Agent Skills Directory

[DATA_EXFILTRATION]: The skill involves accessing sensitive tool-specific files, including the license key at ~/.antibrow/license.key and browser profile data (cookies and local storage) stored within ~/.anti-detect-browser/.
[EXTERNAL_DOWNLOADS]: References external dependencies such as the anti-detect-browser SDK (npm/PyPI) and third-party audit tools like liarjs. It also directs the agent to interact with multiple external fingerprinting analysis services including browserleaks.com, whoer.net, and pixelscan.net.
[COMMAND_EXECUTION]: Instructions include the execution of shell commands such as npx liarjs and python -m antibrow info to inspect system state and verify profile consistency.
[PROMPT_INJECTION]: The skill describes processing data from browser profiles (cookies and local storage), which serves as an ingestion point for untrusted data that could lead to indirect prompt injection.
Ingestion points: Browser profile state directories (~/.anti-detect-browser/<profile>/) containing cookies and session data.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are suggested for when the agent reads or summarizes this profile data.
Capability inventory: The skill uses an SDK capable of launching browser instances, performing network requests, and executing local diagnostic commands.
Sanitization: There is no mention of sanitizing or validating the content of the ingested profile data before it is analyzed or used in logic flows.

multi-account-isolation