The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from external websites into the agent's context, which is an inherent property of web scraping tools.
Ingestion points: Handlers described in references/routing-requests.md and crawler types in references/crawler-choice.md demonstrate how to parse server-rendered HTML and JSON APIs from external URLs.
Capability inventory: The skill possesses the capability to write scraped data to the local file system (Datasets/Key-Value Stores) and initiate further network requests based on ingested content (enqueueLinks).
Boundary markers: The skill lacks explicit instructions for the agent to ignore instructions embedded within scraped content, though it recommends the use of structured data and Zod schemas for validation in references/storage-config.md.
Sanitization: Instructions suggest validating data shapes before storage but do not specifically address sanitizing content for potential prompt injection before LLM processing.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading and installing libraries and browser binaries from well-known registries and official sources.
Evidence: references/runtimes-deploy.md details the installation of dependencies using bun add crawlee and the installation of browser binaries via bunx playwright install chromium.
[COMMAND_EXECUTION]: The skill involves the execution of standard developer CLI tools for running and deploying scrapers.
Evidence: references/runtimes-deploy.md documents usage of apify run, apify push, and standard Node.js/Bun package manager commands.

crawlee