The Agent Skills Directory

[DATA_EXPOSURE]: The skill documentation correctly identifies the risk of handling credentials. It promotes security best practices by recommending a secure Vault for session authentication and the use of environment variables for NOTTE_API_KEY management instead of passing secrets directly through natural language prompts.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to ingest and process untrusted content from external websites, it is inherently susceptible to indirect prompt injection. The documentation provides a mitigation strategy by encouraging the use of Pydantic models for structured extraction, which validates the output against a strict schema and reduces the likelihood of an agent executing instructions embedded in web content.
[COMMAND_EXECUTION]: The skill facilitates browser-level command execution (clicks, form fills, navigation). Recipe 6 demonstrates a feature where the agent generates a Python script from a recorded workflow and writes it to the local filesystem (pricing_extractor.py). This is a documented capability intended for converting exploratory agent runs into repeatable functions.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the notte and notte-mcp packages from PyPI, along with the patchright library for local browser automation. These are official vendor resources or standard technology libraries for this domain.

notte