The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to clone a toolkit from a third-party GitHub repository (github.com/henry-lewiskpp1107/acidrain-security-script-hub.git). This source does not originate from a recognized trusted organization or the skill's primary vendor infrastructure.
[DYNAMIC_EXECUTION]: The toolkit includes JavaScript testing utilities that use new Function(payload) to verify payload execution within various contexts. It also programmatically injects content using element.innerHTML = payload inside test iframes to observe script behavior.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by processing external security testing data and providing capabilities to execute scripts and commands based on that data.
Ingestion points: Loads scripts, configurations, and test cases from the acidrain-security-script-hub repository and local environment variables.
Boundary markers: Lacks technical boundary markers to delimit untrusted test payloads from system instructions, though it includes textual warnings about lab isolation.
Capability inventory: Possesses capabilities for subprocess execution (git clone, php -S, curl) and file system access (logs).
Sanitization: While the skill demonstrates sanitization using DOMPurify and htmlspecialchars, these functions are used as the subjects of security research rather than providing hardened protection for the agent's own execution context.

acidrain-security-testing-toolkit