The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from commit messages and pull/merge request comments, creating a surface for indirect prompt injection.
Ingestion points: Data is ingested from git log (commit messages and bodies), git diff, and pull/merge request metadata and comments via the gh and glab CLI tools (identified in references/create-pr.md, references/read-comments.md, and references/respond-to-comment.md).
Boundary markers: The instructions do not implement specific delimiters or 'ignore' instructions to isolate ingested data from the agent's control flow.
Capability inventory: The skill has access to impactful capabilities including git push, creating pull requests, posting comments, and merging branches (found in references/create-pr.md, references/leave-comment.md, references/merge-pr.md, and references/respond-to-comment.md).
Sanitization: The skill requires explicit human confirmation for every write operation (public or destructive actions) as stated in SKILL.md. It also advises using stdin for passing message bodies to CLI tools to prevent shell command injection.
[COMMAND_EXECUTION]: The skill performs its functions by executing shell commands using git, gh (GitHub CLI), and glab (GitLab CLI).
Evidence: Shell command patterns for branch management, platform detection, and PR lifecycle management are pervasive across references/common.md, references/create-pr.md, references/leave-comment.md, references/merge-pr.md, and references/read-comments.md.

code-pull-request