The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from git logs and diffs while possessing significant repository modification capabilities.
Ingestion points: The skill reads commit messages and file diffs using git log, git show --stat, and git diff as described in the 'Work Flow' section.
Boundary markers: Mandatory plan confirmation ('方案必须先展示给用户确认再执行') and final consistency validation ('git diff <备份引用> HEAD') act as human and automated checkpoints.
Capability inventory: The agent is instructed to use git rebase, git commit, git reset --hard, and git push --force-with-lease.
Sanitization: The skill enforces a specific output format (type(scope): 描述), but does not explicitly sanitize the content retrieved from the history.
[COMMAND_EXECUTION]: The skill executes high-privilege and potentially destructive git commands to modify local and remote history.
Evidence: Explicit instructions to use git reset --hard, git branch -D for cleanup, and git push --force-with-lease for remote updates.
[DYNAMIC_EXECUTION]: The skill involves generating and running dynamic scripts to automate the rebase editor process.
Evidence: The instructions recommend using node -e or shell scripts to programmatically set the GIT_SEQUENCE_EDITOR environment variable for non-interactive rebase operations.

linearize-commits