The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill's instructions for setting up performance testing in GitHub Actions include shell commands that use sudo to register GPG keys, modify system repository sources, and install packages via apt-get. While standard for CI runner environments, these commands facilitate privilege escalation to root during the pipeline execution.
[EXTERNAL_DOWNLOADS]: The provided CI/CD templates reference and download external code from non-trusted community sources, such as the 8398a7/action-slack GitHub Action for Slack notifications.
[REMOTE_CODE_EXECUTION]: The skill facilitates the download and execution of external software and actions during pipeline runs. This includes fetching the k6 load testing tool from official repositories and utilizing the snyk/actions/node security scanning action.
[COMMAND_EXECUTION]: The templates generate extensive shell command sequences using npx, npm, and direct shell invocation to execute linters, test runners (Jest, Playwright), and database migration tools (Prisma).
[INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by interpolating user-provided project data (such as server hosts, ports, and database URLs) directly into shell-executing templates without explicit sanitization logic or boundary markers. Evidence: Ingestion points via placeholders like <app-host> and <port> in SKILL.md; lack of sanitization for these interpolated values; capability for the resulting pipeline to execute arbitrary shell logic based on these parameters.

cicd-pipeline