The Agent Skills Directory

[SAFE]: The skill is entirely instructional and serves to guide the creation of secure CI/CD integrations. It contains no executable scripts, shell commands, or persistence mechanisms.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied information regarding CI environments and tool requirements through a structured interview process.
Ingestion points: User responses to the six-question design interview in SKILL.md and subsequent design discussions.
Boundary markers: The skill does not define internal delimiters for the agent's processing of user input, but it explicitly instructs the agent to output code that uses random delimiters for CI outputs to prevent injection.
Capability inventory: The agent is restricted to generating text-based architectural plans and configuration files; it lacks tools for direct file-system modification or network operations.
Sanitization: The skill provides extensive instructions for sanitizing user-provided values (e.g., Rule 4: quoting variables, Rule 5: SHA pinning) in the code it helps generate.
[SAFE]: All external references and tools mentioned, such as zizmor, actionlint, and cosign, are industry-standard tools for security auditing and signing.
[SAFE]: No obfuscated content, hidden URLs, or unauthorized credential access patterns were found in the skill or its references.

github-action-offering