The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data fetched from external sources, which may contain adversarial instructions.
Ingestion points: Scraped items and job logs are retrieved from Zyte's storage API in references/items-and-logs.md and references/job-validation.md via scripts/scrapy_cloud_api.py.
Boundary markers: The instructions emphasize credential safety but do not define explicit boundary delimiters or warnings to ignore instructions within the scraped data itself.
Capability inventory: The skill possesses the ability to execute shell commands (e.g., shub, uv run), write to local configuration files (e.g., .env, .scrape/.zyte/project-id), and perform network operations to Zyte APIs.
Sanitization: There is no evidence of sanitization, filtering, or escaping of the external content before it is presented to the agent for validation or sampling.
[COMMAND_EXECUTION]: The skill frequently executes local scripts and CLI tools to manage cloud projects and jobs.
Evidence: Multiple references (references/jobs.md, references/deployment.md) utilize uv run and uvx shub for core functionality such as scheduling spiders and deploying code.
[EXTERNAL_DOWNLOADS]: The skill retrieves data and configuration from external services.
Evidence: scripts/stack_tags.py fetches Docker repository tags from a well-known container registry. references/pricing.md uses curl to fetch pricing and domain data from the vendor's official documentation and domain API.

zyte