Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    jujutsu

    1

    Lets coding agents work in Jujutsu (jj) repositories through the jujutsu-mcp tools, without git commands or a shell.

    Free

    1 useSecurity scanned
    jujutsu

    jujutsu

    Example session with this skill installed

    I changed the timeout and added a retry setting in app.py, and started some notes in notes.md. Please commit those as two separate commits, bring my work up to date with what's on origin, and get it ready to push as a branch called feature-x so I can open a PR. I haven't approved the actual push yet.

    • Read your context and instructions
    • Compiled the jujutsu

    Summary

    Your work is two commits on top of the updated origin/main, with the bookmark feature-x on the tip. Nothing has been pushed.

    What happened

    StepToolResult
    Commit app.pycommit (paths: app.py)tswqontr
    Commit notes.mdcommit (paths: notes.md)rvktuxvx
    Update from origingit_fetch, rebase onto trunk()conflict in app.py
    Resolvenew, edit, squash, edit backconflict gone, history straight
    Bookmarkbookmark_set feature-x on rvktuxvxcreated
    Push previewgit_push with dry_runcommits to sign: 2

    Needs your decision

    Conflict on TIMEOUT: upstream already raised it to 60 ("slow networks"); you had 45. I kept 60 plus your RETRIES = 3, and reworded the first commit to feat: add retry setting, since it no longer changes the timeout. Do you want 45 instead?

    debug.log was in your working copy but not in your request. I kept it out of both commits and it is still in @. Should I ignore it, delete it, or leave it?

    Current state

    @  qzlurxkm  (debug.log only, not part of the branch)
    ○  rvktuxvx  docs: start notes on retry policy   feature-x
    ○  tswqontr  feat: add retry setting
    ◆  wopqqtll  fix: raise timeout   main
    

    Next steps

    The push will sign 2 commits, so expect up to 2 security-key touches. Approve it and I'll push feature-x; then open the PR with gh pr create --head feature-x.

    Commit notes.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Coding agents are trained on git. Put one in a Jujutsu (jj) repository and it reaches for git add, opens an editor that never closes, or pushes the wrong commit. This skill teaches the agent to work through the jujutsu-mcp server instead: one typed tool per jj operation, and nothing that can hang waiting for input.

    What it does

    • Detects jj repositories (.jj) and keeps git read-only, with a table that maps each git reflex (add, commit, stash, checkout, pull, push) to the right jj tool.
    • Commits and reshapes history by change id: separate commits, split, squash, rebase, edit an older commit, keeping stray files (logs, build output) out of what gets committed.
    • Syncs with the remote: fetches, rebases onto trunk, resolves conflicts the jj way and returns the history to a straight line. Semantic conflicts (both sides changed the same value) go to you.
    • Prepares pushes safely: points the bookmark at the real tip of the work, runs a dry run first and tells you how many commits will be signed, one security-key touch each.
    • Undoes mistakes through jj's operation log.

    Loads only what the task needs

    SKILL.md is a short router. Each action has its own reference file (commit, reshape, sync, recover, setup, tools), and the agent reads one only when it performs that action.

    Requirements

    • jj 0.45.1 or newer.
    • The jujutsu-mcp server, free and open source (MIT or Apache-2.0): brew install LLawli/tap/jujutsu-mcp or cargo install jujutsu-mcp, then jujutsu-mcp setup to register it in Claude Code, Codex and Antigravity.

    If the server is missing, the skill walks the agent through installing it, asking you first.

    Tested

    Run end to end by an agent on a scenario with a stray file, two separate commits, a conflict with origin and push preparation. It loaded 3 of the 6 references, kept the stray file out, resolved the conflict, left a straight history and stopped at the dry run. That session is the demo on this page.

    Requires the jujutsu-mcp server to be installed and registered. jj commands that open an editor, pager or merge tool are never used; conflicts are resolved by editing files. jj does not push git tags, so tags are created and pushed with git.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Requires

    Jujutsu MCP

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    Shell: `gh pr create`, creating and pushing git tags, and installing the server when you approve it. Read/Write files: reading diffs, resolving conflicted files, editing .gitignore. Network: git fetch and push through jj.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Most AI agents are hardcoded to use Git commands, which can corrupt Jujutsu (jj) repositories or hang indefinitely. This skill acts as a translation layer, teaching agents to use the `jujutsu-mcp` server to handle commits, rebasing, and conflict resolution safely.

    The skill supports any AI agent that can interface with Model Context Protocol (MCP) servers, including Claude Code, Codex, and Antigravity. It requires `jj` version 0.45.1 or newer and the open-source `jujutsu-mcp` server installed on your machine.

    The skill includes a 'SKILL.md' router and six specialized reference modules (commit, reshape, sync, recover, setup, and tools). These modules guide the agent through specific JJ workflows while minimizing token usage by only loading what is necessary for the current task.

    Yes, the skill is designed to handle complex JJ workflows including multi-commit reshaping, rebasing onto trunk, and jj-style conflict resolution. It provides the agent with a mapping table to translate standard Git reflexes into the correct JJ operations like `describe`, `rebase`, and `squash`.

    Safety is built-in: the skill enforces dry runs before any remote push and alerts you to the number of commits requiring signature. It also uses JJ's operation log to allow the agent to undo mistakes, and it prevents the agent from accidentally committing build logs or stray files.

    New in DevOps & Cloud