Git operator

Find the commit that introduced a regression with git bisect, covering a reliable reproduction first, choosing good and bad endpoints, manual and automated runs with a test script, exit codes including skip, narrowing to a path, recovering from a wrong verdict, and always resetting afterwards. Use when a bug appeared between two known points, when a regression must be attributed to a commit, or when a bisect session was left half-finished.

List, create, switch, rename, and delete local Git branches with upstream and unmerged-work safeguards. Use for one focused branch-management operation, not a full pull-request workflow.

Answer quick Git command, flag, and workflow questions with a compact safety-aware reference. Use when the user asks what command to run, what a Git flag means, or which simple operation fits a situation.

Create one reviewed local Git commit from explicitly selected staged changes without pushing or amending by default. Use when the user asks to commit work and wants a clear message and post-commit verification.

Choose, compose, and validate Conventional Commits 1.0.0 messages for task commits created by any commit workflow. Use whenever an agent or user-authorized workflow needs to commit its work, including alongside 11ai-operator-git-commit. This skill governs the message only; it does not authorize staging, committing, amending, pushing, or rewriting history.

Answer questions about the 11ai Git operator plugin: which of its skills fits a task, what each skill promises and refuses to do, which actions always need explicit approval, how staging, committing, syncing, branching, tagging, stashing, recovery, and troubleshooting behave, and where the command and triage references live. Routes every question to the plugin's own contracts and references and answers with a citation. Use when a user asks how the Git operator skills behave, which one to run, or what a skill will and will not touch. For questions about Git itself rather than these skills, use the cheatsheet skill instead.

Write, install, debug, and bypass local Git hooks, covering the hook events and their arguments, why hooks are not committed and how hooksPath changes that, staged-file-only checks, exit codes and blocking behaviour, reading stdin for pre-push and pre-receive, keeping hooks fast, and diagnosing a hook that does not fire. Use when a hook must be added or repaired, when a hook blocks work incorrectly, or when a hook is silently not running.

Connect Git to the systems around it, covering remotes and hosting providers, pull-request tooling, continuous integration triggers, local hooks through husky or lefthook, commit-message and lint gates, large file storage, and submodules or subtrees for vendored code. Use when a repository needs a remote or a second one, when a pipeline trigger does not fire, when a pre-commit gate must be added or is blocking work, or when another repository must be embedded in this one.

Open, inspect, review, and land pull requests from the command line, covering a reviewable branch and description, draft state, reading check results and failing logs, review comments and approvals, keeping a branch current without losing review context, merge strategies and what each does to history, and cleaning up after a merge. Use when a pull request must be created, when its checks or reviews must be read, when review feedback must be addressed, or when it is ready to land.

Recover or undo Git work using restore, revert, reflog, and carefully scoped reset decisions. Use when the user asks to discard local edits, undo a commit, recover a lost tip, or repair a mistaken local Git action.

Install Git and configure identity, default branch name, credential helpers, commit signing with SSH or GPG, global ignore rules, line-ending behaviour, merge and pull defaults, and safe directory ownership, then verify the configuration a commit will actually use. Use when Git is unconfigured on a new machine, when commits carry the wrong author or fail to sign, when pushes prompt for a password, or when the user asks how to set Git up.

Review and stage only the intended Git paths or hunks, or safely unstage them without committing. Use when the user wants to prepare a commit, inspect the index, or remove a path from staging.

Temporarily shelve selected Git work, inspect existing stashes, and apply or drop a stash deliberately. Use when the user explicitly wants to pause work without committing it.

Inspect a Git repository's identity, worktree, diffs, history, branches, and upstream state without changing files or refs. Use when the user asks what changed, whether the tree is clean, where HEAD points, or what should happen next.

Operate Git submodules predictably, covering adding one and what gets committed, cloning and initializing recursively, updating to a pinned commit versus tracking a branch, committing a pointer change, running commands across submodules, removing one completely, and the settings that stop a plain clone or pull leaving them empty or stale. Use when a submodule must be added, updated, pinned, or removed, or when a clone has empty submodule directories.

Compare a local branch with its remote, fetch updates, fast-forward or pull deliberately, and push a named branch when explicitly requested. Use when the user asks to synchronize Git state or resolve ahead and behind status.

Create, sign, publish, and correct Git tags, covering annotated versus lightweight tags, semantic version naming, signing and verification, pushing a single tag rather than all of them, listing and describing the nearest tag, and the rules for retagging a tag that has already been published. Use when a release must be tagged, when a tag must be signed or verified, when a tag points at the wrong commit, or when a tag must be deleted locally and remotely.

Diagnose Git and repository-state failures from reproducible evidence, covering dirty trees, conflicts, detached HEAD, upstream divergence, rejected pushes, remotes, credentials, and recovery. Use when Git reports an error or the repository behaves unexpectedly.

Add, list, move, and remove Git worktrees, covering a second checkout of the same repository, the one-branch-per-worktree rule, detached and new-branch checkouts, shared versus per-worktree configuration, locking a worktree on removable storage, pruning stale administrative entries, and removing a worktree without losing uncommitted work. Use when two branches must be checked out at once, when a worktree cannot be removed or its branch cannot be checked out, or when stale worktree entries must be cleaned up.