pip install git+https://github.com/eelstork/jam.git
Done with jam?
pip uninstall jam
Developer mode: pip install -e . or python install.py after clone.
Opinionated cross platform Git wrapper written mainly for Claude Code (CC) users; not intended for team use.
I use jam land REPO every day; skips the ritualistic git flow that doesn't work when constantly
switching project and mainly retesting features; no need to create a branch in CC web UI.
I do not know how you would benefit or not when using CC in the terminal; I like the web-UI because this provides a clean nix based dev env; then I retest (and test GUIs) on my local, providing a less pristine, bumpier (Windows based) environment.
reclaim (remove CC ads), retro, standup and cooldown (track weekly/daily work) are other commands I do enjoy.
Jam keeps all your repos under one roof (JAM_HOME) and talks to GitHub via gh. No config files, no boilerplate. The stuff you actually do, made fast.
Requires Python 3.10+, git, and gh (authenticated). If you're a github user, of course you already have them installed. Note: I do not believe jam a substitute for knowing and understanding git.
In the doc: 🔺local/remote changes; 🔷local changes only; 🟡informative or config command
Point jam at your repos directory:
jam set-root ~/dev
Or set the JAM_HOME environment variable if you prefer.
🔺jam new NAME ["DESCRIPTION"] |
Spin up a repo on GitHub, clone it locally, push a readme. --public to make it public. Alias: jam create. new.py |
🔺jam copy REPO as NEW |
Copy a repo as a new repo on GitHub. Creates a new private repo, copies all files (minus .git/), commits and pushes. Also: jam copy REPO prompts for the new name interactively. copy.py |
🔺jam rename REPO |
Rename a repo locally and on GitHub. Prompts for the new name. Refuses (without contacting GitHub) if the repo isn't local or if origin points at a different name than the local directory. rename.py |
🔷jam clone NAME |
Clone a repo from your GitHub account into the local jam root. Useful when a repo exists on GitHub but not locally. Also invoked automatically by jam down NAME when the repo isn't found locally. clone.py |
🟡jam list |
See what you've got. --info pulls the first line from each readme. list.py |
🔺jam up "MESSAGE" |
Add everything, commit, push. One shot. --name REPO to target a specific repo, --force if you need it. If the push is rejected because the branch is behind the remote, jam pulls and retries once. up.py |
🔷jam down [NAME] |
Pull latest. If the repo isn't local, delegates to jam clone. --force throws away local changes first. down.py |
🔺jam land [NAME] |
Merge the most recent branch into main and show all landed commits. --all lands across all repos at once. If the push is rejected because main is behind the remote, jam pulls and retries once. land.py |
🔷jam undo [NAME] |
Reverse the last jam command on a repo. Works with up, down, and land. undo.py |
🔷jam delete NAME |
Remove a repo locally. Tags the remote for later cleanup. Re-clone from GitHub to recover. delete.py |
🔷jam prune |
Interactively select and delete repos that only contain a README. Scans jam home, multi-select with arrow keys and x, removes local copies. prune.py |
🔷jam edit FILENAME [REPO] |
Open a file in its default application. If no repo is given, searches all repos. If multiple matches are found, pick interactively. edit.py |
🟡jam tree [NAME] [-L N] |
Show a .gitignore-aware directory tree for a repo. Defaults to depth 2, configurable with -L (matching Unix tree convention). tree.py |
🔺jam remain |
Fix master/main branch confusion across all repos. Installs a Claude Code SessionStart hook that auto-renames local master to main when the remote only has main. --unset removes the hook. Skips repos with uncommitted changes. remain.py |
🔺jam allow-all [REPO] |
Grant Claude Code broad tool permissions across all repos, or just REPO to try it on one first. Merges a permissive permissions.allow list (Bash, Read, Edit, Write, WebFetch, Agent, the GitHub MCP server, and friends) into each repo's .claude/settings.json, then commits and pushes. Existing settings and allow entries are preserved. Per repo: skips unless on main, pulls (--ff-only), edits, commits only that file, then pushes; repos that are off main, behind with conflicts, or carrying unpushed commits are skipped with a one-line reason. --unset removes the rules. allow_all.py |
🟡jam remix |
Itemize and share CLAUDE.md settings across repos. Parses directives from every repo's CLAUDE.md, summarizes and deduplicates them, then lets you interactively share directives to other repos. Non-interactive mode lists all unique directives. remix.py |
🟡jam cooldown |
List today's commits (since 7 am) per repo. Quick end-of-day recap. cooldown.py |
🟡jam stats |
Show command usage counts, most used first. Every command invocation is logged locally to ~/.config/jam/usage.log. --clear discards the log. stats.py |
🟡jam root |
Print the current jam root directory. root.py |
🟡jam set-root PATH |
Set the jam root directory. Writes to ~/.config/jam/root. The JAM_HOME env var takes priority if set. set_root.py |
🟡jam update |
Pull the latest jam source and reinstall. Runs git pull + pip install -e . from wherever jam is installed. update.py |
Repo names can be abbreviated. If the prefix is unambiguous, jam resolves it automatically — jam up -n my-p works if my-project is the only repo starting with my-p. Ambiguous prefixes show matching candidates.
Any command that isn't built-in gets routed to a script at the repo root. Drop a deploy.sh, test.py, or build.ps1 next to your .git and run it with jam deploy, jam test, jam build. Extra arguments are forwarded: jam build --release v2 runs bash build.sh --release v2.
Target a specific repo with jam deploy myrepo — if the second word matches a repo in JAM_HOME, the script runs from that repo. Otherwise it's treated as a regular argument.
Platform-aware: .sh is preferred on Linux/Mac, .ps1 on Windows, .py everywhere. Built-in commands always take priority.
jam l<CMD> REPO [ARGS...] runs jam land REPO first, then the <CMD> passthrough on the same repo. Typical use: jam ldeploy myrepo lands the latest branch and immediately kicks off deploy.sh. If there's nothing to land, the script is skipped and jam exits cleanly; if land fails, the failure propagates and the script is not run. Extra args are forwarded to the script. No --all — run per repo.
Because l is treated as a prefix, scripts whose name starts with l are shadowed by this shortcut.
If you use Claude Code, jam can restore your personal authorship on AI-assisted commits and optionally track coding velocity.
🔺jam claim-commits |
Set up commit attribution. Removes "Claude" from AI-assisted commits so they show your name; commits remain traceable. New repos created with jam new get attribution automatically after setup. claim_commits.py |
🔷jam reclaim [NAME] [-c N] |
Rewrite commit history on the current branch to reclaim authorship on @anthropic.com commits. If a velocity baseline is configured, also tags commits with velocity markers. Use --commits N (or -c N) to limit how far back reclaim looks instead of processing the entire branch. Requires a clean working tree; force pushes the rewritten branch automatically. Don't use this while Claude Code is working on a diff, as it may create (recoverable, but still) confusion. reclaim.py |
🟡jam velocity [NAME] |
Measure coding velocity for a repo. Shows intrinsic (human) and machine-assisted velocity with an acceleration factor. Pick a time period interactively: past week, past month, or all time. velocity_cmd.py |
🟡jam tag-velocity enable/disable |
Toggle velocity tagging on commits. When enabled, jam land and jam reclaim add velocity markers. tag_velocity.py |
🟡jam autofac-reset |
Clear all attribution and velocity config. Removes .claude/ from the current repo and resets claim-commits state so the workflow can be re-run. autofac_reset.py |