Security audit
Security checks for vulnerabilities and agentic risk
This Git workflow skill is mostly coherent, but it includes broad activation language and examples for remote branch deletion and forced history rewrite without strong safeguards.
Install only if you want an agent to help with Git workflow operations and you are prepared to supervise commands that change branches, tags, remotes, or global Git configuration. Treat remote deletion, reset, force-push, and global config examples as manual, confirmation-required actions; review the exact branch list and create backups before running them.
The activation text is very broad, covering generic efficiency, automation, batch processing, and workflow optimization scenarios. In an agent ecosystem with exec/write tools, such vague scope increases the chance this skill is invoked for loosely related requests and may steer the agent toward repository-modifying commands without sufficiently specific user intent.
The manifest summary uses expansive, non-specific phrasing about enterprise workflow support without strong scope boundaries. In practice, that can cause over-selection of the skill and inappropriate application of Git-oriented execution guidance in contexts where the user did not clearly authorize potentially destructive repository actions.
The document provides a pipeline for batch deletion of remote branches using shell filters and git push origin :{} with only a brief '谨慎操作' note, but no mandatory verification or dry-run step. In an exec-capable agent, this can lead to irreversible deletion of shared remote branches at scale if branch matching is wrong, defaults differ, or the agent applies the example automatically.
The rollback guidance includes git reset --hard followed by git push --force-with-lease origin main, which rewrites shared history and can discard collaborators' work. Although the text says to use it cautiously, it lacks a strong, explicit warning and procedural safeguards, making it dangerous in a skill that may be followed or executed verbatim by an agent.
No suspicious patterns detected.
