The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute local git commands and gh (GitHub CLI) operations to analyze repository history and interact with GitHub APIs.
[INDIRECT_PROMPT_INJECTION]: The skill parses untrusted data from commit logs, diffs, and branch names to extract issue numbers and PR metadata, which are then used as arguments in subsequent shell commands.
Ingestion points: Branch names, commit messages via git log, and current repository state (SKILL.md).
Boundary markers: The skill uses a single-quoted shell heredoc (cat <<'EOF') in Step 5 to safely handle the PR body and prevent shell expansion of its content.
Capability inventory: File system access via git, and authenticated network operations via gh api, gh issue view, gh issue edit, and gh pr create (SKILL.md).
Sanitization: While the PR body is handled using safe heredoc practices, the skill does not specify explicit validation or escaping for issue numbers extracted from commit messages before passing them to the gh issue view or gh issue edit commands.

create-pr