The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the official Apify CLI and SDKs from standard registries (NPM, PyPI, Homebrew). It also references official templates and whitepapers from Apify's public GitHub repositories. These are well-known resources from an established technology service.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data as inputs for serverless Actors, creating an attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through Actor.getInput() (JavaScript/TypeScript) or Actor.get_input() (Python) as described in SKILL.md and language-specific reference files.
Boundary markers: The instructions explicitly advise adding boundaries, such as "Validate input types and positive bounded maxItems before execution" and using JSON schemas to define valid input structures.
Capability inventory: The resulting Actors have capabilities including network egress (apify push, web crawling), file system interaction (Key-Value stores, Datasets), and arbitrary command execution within Docker containers.
Sanitization: The skill provides strong sanitization guidance, mandating validation against @apify/json_schemas and recommending the rejection of private-network or credential-bearing URLs.
[COMMAND_EXECUTION]: The skill utilizes the apify CLI tool for project initialization, local testing, and deployment. It also involves creating Dockerfiles and shell wrapper scripts (e.g., start.sh) to execute the actorized software. The instructions include warnings to verify CLI commands and avoid putting sensitive tokens in command history.

apify-actorization