agentleFS
joshuadavidthomasjoshuadavidthomas/agent-skills/jj/SKILL.md

jj

Skill49 starsChanged 6 months agoMITView on GitHub

Jujutsu (jj) — the Git-compatible version control system. Activate ONLY when a .jj/ directory is present in the project or when jj/jujutsu is explicitly mentioned. Do NOT activate for plain git repos without .jj/. Use for any VCS operations in jj-managed projects: commit, push, pull, branch, bookmark, rebase, squash, merge, diff, log, status, working copy, change ID, revset, fileset, template, configuration, workspaces.

  • Commits and pushes

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    The working copy is a commit: No staging area. Every file change is auto-snapshotted into @ when you run any jj command.

    line 29
  2. Distinctive:

    Change IDs are stable. Commit IDs are not: Every commit has two identifiers:

    line 31
  3. Very specific:

    Always use -m for messages: Never invoke a command that opens an editor.

    line 47
  4. Very specific:

    Verify after mutations: Run jj st after squash, abandon, rebase, restore, or any destructive op.

    line 49
  5. Very specific:

    Quote revsets: Always single-quote: jj log -r 'mine() & ::@'.

    line 51
  6. Distinctive:

    When new work doesn't depend on the current chain, branch off trunk:

    line 92
  7. Distinctive:

    Agent rule: Before creating a new commit, decide if it depends on the current chain.

    line 105

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Jujutsu (jj) Version Control

Jujutsu is a Git-compatible VCS with mutable commits, automatic change tracking, and an operation log that makes every action undoable.

Target version: jj 0.36+

Topics

I need to...Deep dive
Understand how jj relates to Git, or use raw git in a jj repogit-compatibility.md
Write revset, fileset, or template expressionsrevsets.md
Push, pull, manage bookmarks, or work with GitHubbookmarks.md
Split, rebase, squash, or resolve conflictsconflicts.md
Run parallel agents with isolated working copiesparallel-agents.md
Configure jj, set up aliases, or customize diffsconfig-reference.md

Mental Model

The working copy is a commit. No staging area. Every file change is auto-snapshotted into @ when you run any jj command. Instead of "stage → commit," just code and describe.

Change IDs are stable. Commit IDs are not. Every commit has two identifiers:

  • Change ID — Stable across rewrites. Letters k–z (e.g., tqpwlqmp). Prefer these.
  • Commit ID — Content hash, changes on any rewrite. Hex digits. This is the Git commit ID in colocated repos.

History is mutable. Commits can be freely rewritten. Descendants auto-rebase. Old versions stay in the operation log.

Bookmarks are not branches. Bookmarks don't advance when new commits are created. They follow rewrites but must be explicitly set before pushing. → Deep dive: bookmarks.md

Conflicts don't block. jj allows committing conflicted files. Resolve at your convenience by editing conflict markers directly, then verify with jj st. → Deep dive: conflicts.md

Agent Rules

Non-negotiable when operating as an automated agent:

  1. Always use -m for messages. Never invoke a command that opens an editor. Commands that need -m: jj new, jj describe, jj commit, jj squash.
  2. Never use interactive commands. jj split (without file paths), jj squash -i, jj resolve — all hang. Use file-path args or jj restore workflows.
  3. Verify after mutations. Run jj st after squash, abandon, rebase, restore, or any destructive op.
  4. Use change IDs, not commit IDs. Change IDs survive rewrites.
  5. Quote revsets. Always single-quote: jj log -r 'mine() & ::@'.

Agent-Specific Configuration

# agent-jj-config.toml
[user]
name = "Agent"
email = "agent@example.com"

[ui]
editor = "TRIED_TO_RUN_AN_INTERACTIVE_EDITOR"
diff-formatter = ":git"
paginate = "never"

Launch with: JJ_CONFIG=/path/to/agent-jj-config.toml <agent-harness> → Deep dive: config-reference.md

Core Workflow

The daily loop: describe → code → new → repeat.

jj describe -m "feat: add user validation"
# make changes — auto-tracked, no `add` needed
jj st && jj diff
jj new -m "feat: add error handling"

Curating History

jj squash -m "feat: final clean message"   # fold working copy into parent
jj absorb                                   # auto-distribute hunks to right ancestor
jj abandon @                               # drop a failed experiment

→ Deep dive: command-gotchas.md

Non-Linear Work

When new work doesn't depend on the current chain, branch off trunk:

# Create sibling from trunk (doesn't move @)
jj new trunk() --no-edit -m "fix: correct timezone handling"
jj edit <bugfix-change-id>
# ... fix the bug ...

# Return to original work
jj log -r 'heads(trunk()..)'
jj edit <feature-change-id>

Agent rule: Before creating a new commit, decide if it depends on the current chain. If not, branch off trunk and flag the divergence to the user.

Pushing Changes

jj bookmark set feat -r @
jj git push -b feat

Bookmarks must be set before pushing — they don't auto-advance. → Deep dive: bookmarks.md

Essential Commands

TaskCommand
Check statusjj st
View diff / logjj diff / jj log
Describe current commitjj describe -m "message"
Start new workjj new -m "task description"
Edit an older commitjj edit <change-id>
Squash into parentjj squash
Auto-distribute changesjj absorb
Abandon a commitjj abandon <change-id>
Undo last operationjj undo
View operation historyjj op log
Restore to earlier statejj op restore <op-id>
Create/move bookmarkjj bookmark create <n> -r @ / jj bookmark set <n> -r @
Push / fetchjj git push -b <bookmark> / jj git fetch

For Git translations: references/git-to-jj.md

Recovery

jj undo                      # undo last op; repeatable
jj op log                    # full operation history
jj op restore <op-id>        # jump to any past state
jj evolog -r <change-id>     # see how a change evolved

Detecting a jj Repo

.jj/ directory = jj repo. Both .jj/ and .git/ = colocated repo. Always use jj commands. Git's "detached HEAD" is normal in colocated repos — use jj log for real state.

Common Mistakes

MistakeFix
Omitting -m on commandsAlways pass -m — editor hangs agents
Using jj split without file pathsProvide paths or use the jj restore workflow
Forgetting to set bookmark before pushjj bookmark set <name> -r @ first
Using commit IDs instead of change IDsChange IDs (letters k–z) survive rewrites
Unquoted revset expressionsAlways single-quote: 'mine() & ::@'
Confusing :: vs .. operators:: = ancestry path, .. = range (see revsets.md)
Creating workspaces as subdirectoriesMust be sibling dirs, not children

Reference Index

Git Interop:

  • references/git-to-jj.md — Git-to-jj command mapping
  • references/git-experts.md — Why jj improves on Git
  • references/git-compatibility.md — Git interop and colocated repos

Commands:

  • references/command-gotchas.md — Flag semantics, quoting, deprecated flags

Revsets & Templates:

  • references/revsets.md — Complete revset language spec
  • references/filesets.md — Complete fileset language spec
  • references/templates.md — Complete template language spec

Sharing:

  • references/bookmarks.md — Complete bookmarks reference
  • references/github.md — GitHub/GitLab workflow details

History:

  • references/conflicts.md — Conflict handling and marker formats
  • references/divergence.md — Divergent changes guide

Config:

  • references/config-reference.md — Full configuration reference

Workspaces:

  • references/parallel-agents.md — Parallel agent setup guide

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.