The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and generate CI/CD configuration files (such as .github/workflows/*.yml and .gitlab-ci.yml) which act as untrusted ingestion points when the agent processes repository content.
Ingestion points: Repository configuration files for GitHub Actions, GitLab CI/CD, Forgejo, Gitea, and Woodpecker as identified in SKILL.md.
Boundary markers: The skill includes a detailed "AI Self-Check" checklist and "AI-Age Considerations" section to help the agent identify and avoid common security pitfalls in generated code.
Capability inventory: Access to a variety of command-line tools for infrastructure management including docker, kubectl, terraform, git, and platform-specific CLIs like gh and glab.
Sanitization: Provides explicit hardening patterns, such as avoiding expression injection in GitHub Actions and utilizing environment variables for secure secret handling, documented in references/github-actions.md and references/supply-chain.md.
[PRIVILEGE_ESCALATION]: Instructions for setting up CI/CD runners in references/runners.md involve necessary system-level operations, such as using sudo for registering runners and installing packages like gitlab-runner.
[REMOTE_CODE_EXECUTION]: The skill provides standard documentation for installing runner environments that involve pipe-to-shell patterns, such as the curl | sudo bash instructions for GitLab runner scripts in references/runners.md.
[EXTERNAL_DOWNLOADS]: Fetches installation scripts and repository configurations from well-known and trusted infrastructure, including GitLab's official package repository (packages.gitlab.com), as part of the infrastructure setup guides.
[COMMAND_EXECUTION]: The skill employs various command-line utilities for pipeline operations, including docker, git, bun, npm, kubectl, and platform CLIs like fj for Forgejo Actions.

ci-cd