Working with Jujutsu Version Control

Working with Jujutsu Version Control

Core Concepts

Jujutsu (jj) - Git-compatible VCS with:

• Change-based: Unique IDs persist through rewrites
• Auto-snapshotting: Working copy snapshotted before each operation
• Stack-based: Build commits in a stack
• Undoable: All ops in jj op log, use jj op restore to time travel

vs Git: No staging area, edit any commit (jj edit), conflicts stored in commits

Working Copy (@)

Current commit is always @:

• jj log -r @ - Current commit
• jj log -r @- - Parent commit
• jj log -r 'ancestors(@, 5)' - Recent stack

State:

• Empty, no description → Ready for changes
• Has changes, no description → Needs description
• Has description + changes → Can stack with jj new
• Has description, no changes → Ready for new work

Stack-Based Workflow

1. Make changes in @ (new files tracked automatically via /jj:commit)
2. Describe: jj describe -m "message" or /jj:commit
3. Stack: jj new
4. Repeat

Why stack: Individual review, easy reordering, incremental shipping, clean history

Plan-Driven Workflow

1. Start: Create "plan:" commit describing intent
2. Work: Implement the plan
3. End: Replace "plan:" with actual work using /jj:commit

TodoWrite: One commit per major todo, jj new between todos

Automatic Snapshotting

Every jj command auto-snapshots the working copy into an unnamed commit. The operation log records every action ever taken:

jj op log              # Full operation history with timestamps
jj op show <op-id>     # What changed in a specific operation
jj undo                # Undo last operation
jj op restore <op-id>  # Time-travel to any point in history

Unlike git reflog, the operation log captures everything: rebases, merges, edits, conflict resolutions — not just branch pointer movements.

First-Class Conflicts

jj stores conflicts inside commits rather than blocking operations. A conflicted state is just another commit you can describe, stack on, or resolve later.

jj rebase -s <source> -d <dest>  # May produce a conflicted commit — doesn't block
jj log -r 'conflicts()'          # List all commits with unresolved conflicts
jj resolve <file>                # Resolve a conflict interactively
jj diff -r <rev>                 # Inspect a conflicted commit's content

In practice: after a rebase that produces conflicts, work continues normally. You can create new commits on top of the conflicted one and resolve it whenever convenient — there is no forced stop like git merge.

Parallel Changes

Create independent lines of work from the same base and rebase them together later:

jj new main -m "feature A"           # New change branching from main
jj new main -m "feature B"           # Another change from same base (parallel)
jj rebase -s <change-b> -d <change-a>  # Stack B on top of A

This is the jj equivalent of working in multiple branches simultaneously, without needing separate worktrees. Conflicts from the rebase are recorded in the commit, not blocking.

When to Suggest Commands

Viewing state: jj status, jj log, jj show, jj diff

Creating commits:

• Use /jj:commit (not jj describe directly)
• Suggest when user has substantial changes or plan needs updating

Organizing commits:

• /jj:split <pattern> when mixing concerns (tests+code)
• /jj:squash for multiple WIP commits
• Don't suggest for simple, focused changes

Undoing: jj undo, jj op restore, jj abandon

Slash Commands

• /jj:commit [message] - Stack commit with message generation
• /jj:split <pattern> - Split by pattern (test, docs, config)
• /jj:squash [revision] - Merge commits
• /jj:cleanup - Remove empty workspaces

Git Submodules

jj does not support git submodules. When working inside a git submodule or managing submodule operations, use git directly:

• git submodule add <url> — add a submodule
• git submodule update --init --recursive — initialize/update submodules
• git -C <submodule-path> <command> — run git commands inside a submodule

Use jj only for the parent repository (non-submodule paths).

Git Translation

Repository blocks git write commands via hook. Prefer jj equivalents:

• git status → jj status
• git commit → /jj:commit
• git log → jj log
• git checkout → jj new

jj restore — Use With Care

jj restore --from <rev> -- <path> replaces the target path's content with the state from <rev>. Any file present in --to but absent in --from will be deleted. This has caused accidental mass-deletions when the source revision didn't include recently added files.

Before running jj restore:

1. Check what exists in the source: jj diff -r <rev> -- <path>
2. Check what would change: jj diff --from <rev> -- <path> (dry-run view)
3. After restoring, run jj diff --summary to verify no unexpected deletions

Best Practices

Do: Stack commits, describe clearly (what/why), use plan-driven workflow, leverage jj op log, split mixed concerns

Don't: Mix git/jj (except for submodules), leave work undescribed, create monolithic commits, forget everything is undoable