The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: In references/cli-actorization.md, a Dockerfile example includes the command curl ... | sh to install the ubi utility from the houseabsolute/ubi repository. This piped remote execution pattern is a security risk as it bypasses integrity checks, a risk explicitly noted in the skill's own SKILL.md file.- [DATA_EXFILTRATION]: The skill provides instructions on handling APIFY_TOKEN credentials. It correctly advises against passing tokens as command-line arguments or embedding them in source code, recommending environment variables or interactive login instead.- [EXTERNAL_DOWNLOADS]: The skill references various external resources, including the apify-cli (via npm), the Apify SDK, and official Apify GitHub repositories. These are legitimate resources for the vendor.- [PROMPT_INJECTION]: While the skill doesn't contain malicious injections, it acknowledges the risk of indirect prompt injection from crawled web content. It provides mandatory security guidelines for sanitizing and validating external data. Mandatory Evidence Chain: (1) Ingestion points: Actor.getInput() and web crawlers; (2) Boundary markers: Present as instructions in SKILL.md; (3) Capability inventory: Subprocess execution in shell wrappers and SDK network operations; (4) Sanitization: Explicitly recommended in the Security section.

apify-actorization