Claude-skill-registry jj-workflow

Jujutsu (jj) version control, load skill when hook output shows vcs=jj-colocated or vcs=jj in the system-reminder.

install
source · Clone the upstream repo
git clone https://github.com/majiayu000/claude-skill-registry
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/jj-workflow" ~/.claude/skills/majiayu000-claude-skill-registry-jj-workflow && rm -rf "$T"
manifest: skills/data/jj-workflow/SKILL.md
source content

jj Workflow

CRITICAL: Avoid Interactive Mode

Always use

-m
flag to prevent jj from opening an editor:

# WRONG - opens editor, blocks AI
jj new
jj describe
jj squash

# CORRECT - non-interactive
jj new -m "message"
jj describe -m "message"
jj squash -m "message"

Never use these interactive commands:

  • jj split
    - inherently interactive, no non-interactive mode

Mental Model

No staging area. Your working directory is always a commit. Every save is tracked.

  • @
    = your current change (the working copy)
  • @-
    = parent of current change
  • Changes are mutable until pushed

When to Use What

SituationDo This
Starting new work
jj new -m "what I'm trying"
Forgot to start with jj new
jj describe -m "what I'm doing"
(do this immediately)
Work is done, move on
jj new -m "next task"
Annotate what you did
jj describe -m "feat: auth"
Broke something
jj op log
jj op restore <id>
Undo one file
jj restore --from @- <path>
Combine messy commits
jj squash -m "combined message"
Try something risky
jj new -m "experiment"
, then
jj abandon @
if it fails

AI Coding Pattern

Always have a description. The working copy should never stay "(no description set)".

# BEFORE starting work - declare intent
jj new -m "feat: add user logout button"
# Now implement... jj tracks everything automatically

# FORGOT to start with jj new? Describe immediately
jj describe -m "feat: what I'm working on"

Why this matters:

  • jj log
    shows meaningful history while working
  • Easier to understand what each change does
  • Simpler to curate/squash later
  • Teammates can follow progress
# Checkpoint before risky changes
jj describe -m "checkpoint: auth works"
jj new -m "trying OAuth integration"

# If it breaks
jj op log              # Find good state
jj op restore <id>     # Go back

# When done, curate history
jj squash -m "feat: OAuth support"

Push to GitHub

Pushed commits are immutable. You can't squash into or modify them. The safe pattern:

# 1. Abandon empty checkpoint commits cluttering history
jj log -r '::@'                      # Find checkpoints
jj abandon <change-ids>              # Remove empty ones

# 2. Describe your work (don't try to squash into immutable parent)
jj describe -m "feat: what you did"

# 3. Move bookmark to your commit and push
jj bookmark set master -r @
jj git push

For feature branches (new):

jj bookmark create feature-x -r @
jj git push --bookmark feature-x
# If refused, configure auto-tracking once:
jj config set --user 'remotes.origin.auto-track-bookmarks' 'glob:*'
# Then retry: jj git push --bookmark feature-x

For feature branches (updating):

jj bookmark set feature-x -r @
jj git push

Teammates see clean git. They don't know you used jj.

Recovery

The oplog records every operation. Nothing is lost.

jj op log                      # See all operations
jj undo                        # Undo last operation
jj op restore <id>             # Jump to any past state

Bail Out

rm -rf .jj    # Delete jj, keep git unchanged