apify-actorization
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides a Dockerfile template for CLI-based actors that downloads and executes a shell script directly from a remote source.
- Evidence: In
references/cli-actorization.md, a commandcurl --silent --location https://raw.githubusercontent.com/houseabsolute/ubi/master/bootstrap/bootstrap-ubi.sh | shis used to install a utility. While this is a common pattern for the recommended tool, piping remote scripts to a shell bypasses integrity checks and is a known security risk. - The skill also suggests using
npx @playwright/mcp@latest, which executes code directly from the npm registry. - [EXTERNAL_DOWNLOADS]: The skill instructs the user to download several external packages and tools during setup.
- Evidence: Instructions in
SKILL.mdincludenpm install -g apify-cliandpip install apify. - Reference files recommend installing
jqandubivia external repositories. - [COMMAND_EXECUTION]: The skill involves the execution of various shell commands to manage the Actor lifecycle.
- Evidence: Commands such as
apify init,apify run,apify push, andapify loginare central to the skill's functionality. - The skill provides guidance on safely passing inputs to these commands, specifically warning against passing tokens as command-line arguments to avoid exposure in process listings or history.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies a significant attack surface regarding the processing of external data.
- Ingestion Points: Actors are designed to crawl and ingest data from external websites (
SKILL.md). - Capability Inventory: The generated actors have capabilities for network operations, file system writes (datasets/key-value stores), and potential shell command execution in wrapper scripts.
- Sanitization: The skill includes a dedicated 'Security' section in
SKILL.mdthat explicitly warns users to sanitize crawled data, validate types, and avoid treating scraped text as code or commands to prevent prompt injection and other attacks. - Boundary Markers: The skill encourages the use of structured JSON schemas for input and output to define strict data boundaries.
Audit Metadata