The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the standard GitHub CLI (gh) to perform read-only operations such as checking PR status (gh pr checks), viewing PR details (gh pr view), and listing changed files (gh pr diff). These are standard development operations.
[DATA_EXPOSURE]: The instructions mention environment variables like CODECOV_TOKEN and MKIT_RELEASE_ATTEST_KEY in the context of documentation for CI/CD configuration and troubleshooting. There are no instructions to exfiltrate these keys or access sensitive local files (e.g., SSH keys).
[INDIRECT_PROMPT_INJECTION]: The skill involves reading external data from the GitHub API and repository files (e.g., .github/workflows/*.yml, cloudbuild/*.yaml).
Ingestion points: Output from gh commands and local repository configuration files.
Boundary markers: None explicitly defined, as the skill provides a workflow for the agent to follow based on observed data.
Capability inventory: The skill is restricted to informational commands and does not perform file writes, remote code execution, or network exfiltration based on the ingested data.
Sanitization: Not applicable as the skill focuses on status verification and does not process untrusted data into executable contexts.
[METADATA_POISONING]: The skill's metadata and description accurately reflect its purpose as a CI preflight checklist. No deceptive instructions were found in the frontmatter.

mkit-ci-preflight