The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites via notte page scrape and notte sessions network. This content is then used by the agent to generate and edit executable Python code (built_function.py).
Ingestion points: Web content is fetched in SKILL.md and references/exploration.md using session-based scraping tools.
Boundary markers: The skill provides manual verification steps in the 'Security' section of SKILL.md, advising the agent to ensure exploration reached the intended data rather than lookalike content.
Capability inventory: The agent has access to Bash (including notte and curl), Write, and Edit tools, as well as the ability to deploy and execute code via notte functions create and notte functions run.
Sanitization: The skill instructs the agent to 'Clean the export before relying on it' in Phase 3 of SKILL.md, specifically fixing syntax errors and escaping issues.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI tools via the Bash toolset. This includes notte commands for browser control, cloud function management, and curl for HTTP interactions. It also uses notte page eval-js to execute arbitrary JavaScript in the context of the target website for data discovery.
[DYNAMIC_EXECUTION]: The core functionality involves generating a Python script (built_function.py) from a recorded browser session using notte sessions workflow-code. This generated code is then modified by the agent and executed in a cloud environment using notte functions run.

notte-functions-build