The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute several shell and Git commands to perform its core function. These include generating a branch ID using date and sha1sum, creating new branches with git checkout -b, committing code, and pushing to remote repositories via git push. Additionally, it performs cleanup of temporary state files in ~/.claude/cooking/ using rm -f. These actions are consistent with the documented purpose of multi-repo refactoring.
[PROMPT_INJECTION]: The skill defines a structured command surface where it ingests a user-provided "Description" and a list of repositories. This input is then used to direct subagent tasks, creating a surface for potential indirect prompt injection if the user-provided description contains malicious instructions for the subagents.
Ingestion points: The Slug and Description parameters provided by the user in the execution prompt (SKILL.md).
Boundary markers: The instructions do not define specific delimiters (like XML tags or tripple-backticks) to isolate the user-provided description when passing it to subagents, though it does mandate a structured commit message format.
Capability inventory: The skill has the ability to modify source code, commit changes, and push code to remote origin repositories (SKILL.md).
Sanitization: The skill relies on an external shell script (branch-guard.sh) to validate branch naming conventions, but does not specify runtime sanitization for the refactoring description itself.

cross-repo-refactor