The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: In references/cli-actorization.md, the skill provides a Dockerfile template that instructs the user to download a bootstrap script from a third-party GitHub repository (houseabsolute/ubi) and pipe it directly to a shell for execution (curl | sh). This pattern is a significant security risk as it executes unverified code from an external source.
[EXTERNAL_DOWNLOADS]: The skill instructs users to download several packages and tools: it fetches the official Apify CLI from the npm registry, the official Apify Python SDK from PyPI, and a utility binary from a third-party GitHub repository.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to create applications that ingest and process untrusted data from external websites, which creates a vulnerability surface for indirect prompt injection. 1. Ingestion points: Data is retrieved via Actor.getInput() (references/js-ts-actorization.md), Actor.get_input() (references/python-actorization.md), and apify actor:get-input (references/cli-actorization.md). 2. Boundary markers: No boundary markers or delimiters are present in the provided templates to isolate untrusted data. 3. Capability inventory: The resulting Actors possess network access, storage access, and potential shell execution capabilities. 4. Sanitization: The skill contains documentation warnings about sanitizing crawled data, but no automated sanitization logic or validation libraries are included in the provided code snippets.
[COMMAND_EXECUTION]: The CLI wrapper template (start.sh) executes local application binaries using parameters derived from external JSON input (apify actor:get-input), which could lead to command injection if the input is not strictly validated.

apify-actorization