The Agent Skills Directory

[SAFE]: The skill implements a robust security model for repository interaction. It mandates the use of authenticated official CLI tools (gh for GitHub, fj for Forgejo) and prevents silent mutations by requiring explicit user intent for merge operations.
[COMMAND_EXECUTION]: The skill uses standard Git and forge CLI commands (e.g., git push, gh pr create) which are necessary for its primary purpose of PR management. It includes safety flags like --force-with-lease and --match-head-commit to prevent accidental overwrites.
[DATA_EXPOSURE]: The workflow correctly instructs the user to handle Forgejo tokens via interactive commands rather than shell arguments or files, and explicitly forbids the skill from reading or reporting these tokens.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risk of processing external content (PR templates and existing bodies) and provides instructions to preserve reviewer context while removing unfilled placeholders, effectively managing the ingestion surface.

draft-pr