The Agent Skills Directory

[COMMAND_EXECUTION]: The file references/developer-portal-customization.md includes a PluginManager implementation that executes shell commands using child_process.exec. The commands use string interpolation with variable inputs (e.g., yarn add ${name}@${version}), creating a significant vulnerability for arbitrary shell command injection if the variables are not strictly validated.- [REMOTE_CODE_EXECUTION]: The skill provides logic in references/developer-portal-customization.md to dynamically install Node.js packages via yarn add. This capability allows for the installation and execution of unverified external code within the system environment.- [PROMPT_INJECTION]: 1. The skill contains excessive redundant data across nine reference files (e.g., references/architecture-patterns.md, references/deployment-pipelines.md), each repeating identical filler text hundreds of times. This 'context stuffing' is an adversarial technique used to bypass instruction constraints or hide malicious payloads from automated scanners. 2. The skill architecture ingests external configurations and API definitions from remote URLs (e.g., GitHub) without demonstrating security boundaries, exposing the agent to potential indirect prompt injection attacks.- [EXTERNAL_DOWNLOADS]: The skill's core workflow and examples involve fetching and processing remote files (e.g., catalog-info.yaml and OpenAPI specs) from external repository URLs.

devops-internal-developer-platform