The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: In references/cli-actorization.md, the provided Dockerfile template executes a remote bootstrap script using curl | sh (https://raw.githubusercontent.com/houseabsolute/ubi/master/bootstrap/bootstrap-ubi.sh | sh). This pattern executes unverified code directly from the internet, which can be exploited if the source repository or the delivery mechanism is compromised. This is particularly notable as SKILL.md explicitly warns against this exact practice for the apify CLI.
[EXTERNAL_DOWNLOADS]: The skill recommends downloading and installing the ubi tool from the houseabsolute GitHub repository. This repository does not belong to a pre-verified trusted organization or well-known service, increasing the risk of supply chain attacks.
[COMMAND_EXECUTION]: The wrapper script example in references/cli-actorization.md (start.sh) demonstrates parsing JSON input and using it directly in command-line arguments (./your-application --param "$MY_PARAM"). While the example uses jq and quotes, it establishes a pattern where unsanitized external input (crawled data) could lead to command injection if not handled with extreme care by the developer.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to process data from external websites, which is a major ingestion point for untrusted content.
Ingestion points: Crawled web content, titles, and HTML are ingested in references/js-ts-actorization.md and references/python-actorization.md.
Boundary markers: The skill mentions that users should sanitize crawled data but does not provide automated boundary markers in its code templates.
Capability inventory: The resulting Actors have capabilities for network operations, file system access via key-value stores, and database writes.
Sanitization: The SKILL.md file includes a 'Security' section advising developers to sanitize data, use parameterized APIs, and avoid executing crawled content, which mitigates the risk but the surface remains inherent to the skill's function.

apify-actorization