The Agent Skills Directory

Privilege Escalation (HIGH): The file references/testing-patterns.md contains templates using sudo for package installation (sudo apt-get install). While common in CI/CD environments, the use of sudo represents privilege escalation within the execution context.
Unverifiable Dependencies & Remote Code Execution (MEDIUM): The skill templates initiate external downloads using system package managers (apt-get, dnf, brew) and third-party GitHub Actions like goreleaser/goreleaser-action. These are external dependencies that are not part of the trusted source list.
Indirect Prompt Injection (HIGH): The skill's primary purpose is to troubleshoot and modify .github/workflows files based on user-provided data. This creates a high-risk surface for indirect prompt injection where malicious code or logs could influence the agent to suggest harmful pipeline modifications.
Ingestion points: Processes user-provided .github/workflows files and CI error logs via the skill description instructions.
Boundary markers: None present; the templates do not define delimiters for untrusted content.
Capability inventory: Capable of generating shell commands, suggesting code modifications, and configuring repository permissions/secrets.
Sanitization: No validation or sanitization logic is provided to verify the integrity of the code being analyzed.
Unverifiable Dependencies & Remote Code Execution (LOW): References several trusted GitHub Actions from the actions/ organization (e.g., checkout, setup-go, cache) which are downgraded to LOW per the trusted source rule.
Command Execution (LOW): Includes standard shell patterns for local binary execution (./dotfiles-installer), environment configuration (export HOME), and testing operations (go test).

configuring-github-actions