The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation includes an advisory warning users about the risks of indirect prompt injection from untrusted web pages. It identifies that content from notte page scrape or notte agents start could contain malicious instructions.
Ingestion points: Data enters the agent context through notte page scrape, notte page observe, and notte page eval-js (SKILL.md, references/session-management.md).
Boundary markers: The CLI does not automatically apply boundary markers to scraped content, though the documentation advises users to include them in their extraction instructions.
Capability inventory: The skill allows full page interaction including click, fill, goto, and eval-js (SKILL.md).
Sanitization: No automatic sanitization of web content is performed; the skill relies on the caller to validate scraped data before further use.
[CREDENTIALS_UNSAFE]: The skill provides a vault system and persona management for handling authentication. The documentation follows security best practices by explicitly warning against passing secrets as command-line arguments and recommending the use of environment variables or the built-in vault to prevent credential leakage in shell history or process snapshots.
[EXTERNAL_DOWNLOADS]: Setup instructions involve installing the CLI via Homebrew or Go from official vendor repositories (github.com/nottelabs/notte-cli). These resources are hosted by the skill's author and are used for standard installation.
[COMMAND_EXECUTION]: The skill enables browser automation, which involves executing JavaScript within browser sessions (notte page eval-js) and deploying Python-based workflows (notte functions). These are core, documented functionalities for the intended use-case of web automation and testing.

notte-browser