The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions define shell command patterns that incorporate user-supplied inputs, creating a potential command injection surface. Specific examples include:
Interpolation of commit messages in git commit --amend -m "<新メッセージ>" and git commit -m "<message>".
Interpolation of branch/revision names in git rebase --onto <newbase> <upstream>.
Execution of shell scripts via the GIT_SEQUENCE_EDITOR and GIT_EDITOR environment variables using sh -c, which facilitates dynamic shell execution to bypass interactive prompts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository which could contain malicious instructions designed to influence agent behavior.
Ingestion points: The agent reads commit logs (git log), file differences (git show), and file contents during conflict resolution (SKILL.md).
Boundary markers: There are no explicit instructions for the agent to ignore or delimit potentially malicious instructions found within commit messages or file diffs.
Capability inventory: The skill possesses the ability to write files, modify the repository state, and execute shell commands through the git editor mechanism.
Sanitization: The skill does not specify any sanitization or escaping protocols for repository-derived data before it is processed by the agent.

git-rebase