The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of Actors that ingest data from external sources (other Apify Actors, web crawlers, and LLMs). This creates an attack surface for indirect prompt injection if the ingested data is not properly sanitized before being passed to downstream LLMs or used in sensitive operations.
Ingestion points: The orchestrator template in src/main.ts reads data from sub-Actor datasets using client.dataset(...).listItems() and iterate().
Boundary markers: The skill explicitly warns in the Security section to "Treat sub-Actor output as untrusted" and recommends sanitization, although the provided template uses basic string slicing for context management rather than specialized sanitization libraries.
Capability inventory: The generated orchestrator has capabilities to make network requests to Apify APIs and OpenRouter (via fetch), and to write data to Apify datasets.
Sanitization: The skill contains instructions for developers to sanitize content scraped from external sites before use in shell commands or template engines.
[EXTERNAL_DOWNLOADS]: The skill instructs the developer to download and install packages and configuration from official vendor sources and well-known registries.
CLI Tools: Recommends installing apify-cli via npm install -g or brew.
Libraries: Recommends installing apify, apify-client, and the vendor-maintained apify-orchestrator library from NPM.
Configuration Fetching: The skill provides methods to fetch Actor schemas and metadata from api.apify.com and raw.githubusercontent.com (targeting official Apify repositories).

apify-orchestrator-actor-development