The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses uv run python to execute a local script scripts/provider_oracles.py which performs surface discovery, result recording, and browser automation.
[EXTERNAL_DOWNLOADS]: The skill interacts with multiple external services including OpenAI APIs, Microsoft Azure Blob Storage, and residential proxy services (ThorData). These interactions involve uploading media artifacts for watermark analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media files (image, video, audio) and interprets the textual output from external provider interfaces.
Ingestion points: External media files are read from the uploads/ directory defined in a manifest.json file.
Boundary markers: The instructions mandate action-time confirmation before each upload and require the agent to record verbatim responses rather than interpreting them.
Capability inventory: The skill has the capability to execute shell commands, automate browsers (Chrome and Playwright), and perform authenticated network requests to multiple cloud providers.
Sanitization: The skill implements slot validation and profile matching for browser sessions to ensure actions occur within the intended context.
[CREDENTIALS_UNSAFE]: While the skill does not hardcode secrets, it references a large number of sensitive environment variables for API keys (OPENAI_API_KEY_N), Azure endpoints, and residential proxy URLs (THORDATA_ROUTE_*) which must be managed in a gitignored .env file.

provider-oracles