The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied CI/CD configuration files (e.g., Jenkinsfile, GitLab-CI) to generate new workflows, creating a surface for embedded instructions to influence the agent.
Ingestion points: Phase 1 (Source) in SKILL.md requires actual source CI/CD files from the user.
Boundary markers: The skill includes explicit constraints to "Convert only what's in the source" and "Never invent workflows from descriptions."
Capability inventory: The skill utilizes shell commands for file management (mv, rm), repository metadata retrieval (gh api), and workflow validation (actionlint).
Sanitization: No specific sanitization or filtering of the source file content is described before processing.
[COMMAND_EXECUTION]: The skill instructs the agent to perform several shell-based operations:
Archival Protocol: Uses mkdir, mv, and rm to move original source files to a .github/ci-archive/ directory and delete them from their original locations.
Version Resolution: Uses gh api to fetch release tags and commit SHAs from GitHub repositories when resolving action versions.
Validation: Executes actionlint to validate the generated workflows.

migration-core