The Agent Skills Directory

[SAFE]: The skill establishes a high standard for data privacy by requiring that all identity evidence be sanitized and non-reversible. The shared browser-operation protocol explicitly forbids storing sensitive identifiers such as email addresses, cookies, tokens, or raw authentication secrets, mandating the use of one-way hashes and minimal direct evidence labels instead.
[SAFE]: The skill contains comprehensive instructions to mitigate risks associated with untrusted data ingestion. Ingestion points: Webpage DOM, accessibility trees, console logs, and network traffic from targeted URLs. Boundary markers: The protocol utilizes a formal Handoff Request system with unique operation_id trackers and explicit authorization scopes to ensure actions are vetted. Capability inventory: Navigating pages, DOM inspection, screenshot capture, form filling, and file upload/download. Sanitization: Instructions mandate treating all page content as untrusted input, ignoring instructions found within processed data that conflict with user tasks, and reporting suspected injection attempts.
[SAFE]: The workflow design prevents session confusion and unauthorized actions by requiring explicit re-verification of the target origin and login state before performing any sensitive operation, such as submitting forms or uploading files. It strictly isolates browser surfaces (e.g., cloud vs. desktop) to prevent unintended state leakage.

ops-browser