GitInProgress
Description
AI drafts the commits, you decide what ships. GitInProgress lets your AI read your uncommitted working tree and propose how to split it into focused commit candidates — each with a draft commit message. Nothing lands in git history until you review and approve; the AI never…
About
AI drafts the commits, you decide what ships. GitInProgress lets your AI read your uncommitted working tree and propose how to split it into focused commit candidates — each with a draft commit message. Nothing lands in git history until you review and approve; the AI never commits on its own.
Details
- Author
- Unknown
- Categories
- Developer Tools, Other, Automation
Jump to
AI drafts the commits, you decide what ships. GitInProgress lets your AI read your uncommitted working tree and propose how to split it into focused commit candidates — each with a draft commit message. Nothing lands in git history until you review and approve; the AI never commits on its own.
Group your AI's changes into focused commit candidates — edit the draft messages; nothing lands until you approve.
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
AI drafts the commits. You decide what ships.GitInProgress turns a messy working tree intoreviewable commit candidates: your own AI proposes how to split the changes into focused commits — each with a draft commit message — thenyourefine the grouping, tweak the messages, and commit.Nothing lands in history until you approve it— the AI never commits on its own. Along the way you can test each grouping in isolation and iterate safely.
Git records commitsafter the fact. GitInProgress shapes thembefore— nothing lands in history until you approve.
Ask your AI"Draft my commits"— GitInProgress splits a messy working tree into focusedcommit candidates, each with a draft message. You review the grouping, stage them, and commit;nothing lands in git history until you approve. (Opens in thebottom panel, alongside the integrated Terminal.)
Stop staring at a giant pile of unsorted changes. Your own AI client (via MCP or GitHub Copilot) reads your working tree and proposes a clean way to break it intocommit candidates— small, focused groups of changes, each with a ready-to-edit commit message drafted for you. Just ask your AI:"Draft my commits."
- You stay in control— these are proposals, not commits. Nothing enters your git history until you review, tweak, and confirm.
- Nothing slips through the cracks— changes you haven't sorted yet collect in anInbox, pinned to the top, so you always know what still needs a home. You or the AI file them into candidates from there.
- Organize the way you think— gather changes into named candidates (we call themChangeBoxes) with drag & drop, multi-select, color-coding, and reordering.
- Split even within one file— a single file often touches two unrelated things, so a clean split can't stop at file boundaries. Ask your AI to assign each individual edit (a "hunk") to a different ChangeBox — e.g. "put the debug logging in Debug." The commit then stages only the lines that candidate owns; the rest stay untouched.
- Commit one group at a time— turn a single candidate into a commit without disturbing anything else in your working tree.
Each candidate (aChangeBox) groups related files with an editable draft message — review it, tweak it, and commit one at a time. TheDebugbox is parked, so it stays out of every commit.
Safety nets ⚡ Pro Preview (currently free)
Experiment freely — you can always get back to a known-good state. (On the Free tier, deleting a file simply discards it, with no recovery.)
- Archive— a non-destructive checkpoint that preserves a candidate's ChangeBox grouping while clearing it out of your working tree. Restore it by hand whenever you need it.
- Snapshot— capture a restorable copy of your workwhile leaving it in place, so you can keep editing.
- Auto set-aside on branch switch / pull— switching branches or pulling with uncommitted work automatically sets it aside (grouping intact) and offers to bring it back when you return.
- Trash— a recycle bin with a 30-day recovery window. Restore single files or whole ChangeBox groups. Trash is only ever cleared when you explicitly choose to.
- Trash-backed AI iteration— a candidate that didn't pan out goes to Trash, where you can ask the AI for a different take without losing the old one.
Archive lifts your work out of the working tree while keeping each candidate's grouping — parked boxes included — ready to restore by hand whenever you need it.
Trash keeps removed files for 30 days, grouped by the candidate they came from — restore a single file or a whole group, and it's cleared only when you say so.
Isolation testing ⚡ Pro Preview (currently free)
Try a slice of your work on its own to see if it actually builds and passes — without losing the rest.
- Test a prefix in isolation— choose how many of your commit candidates to keep live — the first N, as a prefix (#1,#1–#2, …) — build/test just those, and the rest are set aside, then restored when you're done.
Keep a prefix of your candidates (here#1–#2) live to build and test on their own; everything below the line is set aside, then restored when you're done.
Remember why you were here and where you left off.
- Branch memos— attach a note and a status (In Progress / Review / Waiting Merge / Abandoned) to each branch. When you switch back, the previous note pops up so you instantly pick up the thread.
- File progress— see each file'sunstaged and staged changes side by sideon one row, with added / removed line counts and arrow buttons to stage or unstage, in a collapsible tree (or flat list).
Finish the job without leaving the panel — commit and push with automatic remote detection and warnings when your branch isn't tracking a remote.
Let your AI committhe right way. GitInProgress ships anMCP (Model Context Protocol) server(stdio) so AI clients — Claude Code, GitHub Copilot Chat, Cursor, and any MCP-compatible client — can read your work-in-progress and committhroughyour candidates instead of running rawgit commit. It exposes read tools plus write tools that are always recoverable; destructive actions (emptying trash, permanent delete) aredeliberately left out, so an AI can never wipe your work.
→ SeeMCP_SETUP.mdfor client registration steps and the full tool list.
GitInProgress isfree for everyone during early access.
- AI commit splitting (create commit candidates from your own AI)
- Manual ChangeBox editing, per-changeBox commit, and Commit All
- Inbox / Branch memo — deleting a file discards it (no recovery)
- MCP server integration (Claude Code / Copilot / Cursor read & commit through your splits)
- Safety nets— Archive, Snapshot, Auto set-aside on branch switch / pull, and Trash (30-day recovery; on Free, deleting a file just discards it)
- Isolation testing— build/test the first N commit candidates (a prefix) on their own
- Trash-backed AI iteration— failed candidates go to Trash; ask the AI for an alternative
Pro pricing isTBD, planned for thev1.0+ stablerelease. During early access, Pro Preview features are free so you can evaluate the full feature set.
All data (memos, file progress, change boxes, trash metadata) is saved under.git/gitinprogress/inside your repository. It'slocal to each repo and ignored by git— nothing is sent to any server. Your AI client only sees this data whenyouinvoke the MCP server.
For worktrees, each worktree gets an independent state at<main>/.git/worktrees/<name>/gitinprogress/(seedocs/WORKTREE_SUPPORT.mdfor details).
- VS Code 1.105+(the bundled MCP server integration needs 1.99+)
- A Git repository in the workspace
- Single-root workspace only: multi-root workspaces are not supported in v0.1.x — only the first workspace folder is tracked. Open each project in its own window for now.
- Forgit worktreeusers: open each worktree directory in a separate window (worktree-aware: ✅, cross-worktree view: ❌).
- Install the extension from the Marketplace.
- Open a git repository in VS Code.
- Open thebottom panel(Ctrl/Cmd+J) and select theGitInProgresstab — it sits next to Terminal, Output, and Problems.
- (Optional) Connect your AI client — seeMCP_SETUP.md.
- After editing files, ask your AI:"Draft my commits"— then review and commit each candidate yourself. (No AI connected? Group changes by drag & drop instead.)
GitInProgress runs in thebottom panel, like the integrated terminal — open it withCtrl/Cmd+Jand pick theGitInProgresstab (next to Terminal / Output / Problems).
GitInProgress is maintained by a single developer (solo dev). Bug reports and feature requests are triaged on abest-effort basis— there is no SLA. During early access, you're encouraged to use the Pro Preview features and share feedback before the v1.0 launch.
Open an issuein the public issue tracker. Please use the provided issue templates and include:
- VS Code version
- GitInProgress version
- OS
- Steps to reproduce
- Expected vs actual behavior
- Logs from theOutput → GitInProgresschannel (if relevant)
Issues without sufficient reproduction information may be closed after 7 days of inactivity.
Do not file public issues for security concerns.Please use GitHub'sprivate security reporting:
- Go to theSecurity tabof the issue tracker
- Click "Report a vulnerability"
- Fill in the form with details
You should receive an acknowledgment within 7 days.
Use the issue tracker with thequestionlabel.
A dedicated email channel for Pro tier users will be added when the Pro tier launches in v1.0+.
GitInProgress isproprietary software— seeLICENSEfor the full terms. The source is not publicly distributed; copying, modification, redistribution, and reverse engineering are not permitted. During early access (v0.1.x) the extension is free; some features may require a paid Pro license from v1.0 onward. The early-access period is your chance to evaluate everything first.
This extension bundles third-party open-source components, each under its own permissive license (MIT / ISC / BSD). SeeTHIRD_PARTY_NOTICES.mdfor the full list and license texts.
DOS is a small open-source kernel that doesn't believe your AI agents: it verifies what actually shipped from git evidence, arbitrates file collisions between parallel agents, and refuses with structured reasons.
Make git commits on behalf of AI to track AI contributions in your codebase.
An MCP server for interacting with and automating Git repositories using Large Language Models.
Generates Pull Request messages based on Git commits and changes.
Anchor is local repo and org memory for AI coding agents. It indexes GitHub PR history, current code, tests, regressions, architecture, and cross-repo impact locally, then exposes concise cited context through MCP and CLI workflows. Local-first. Read-only GitHub access. No CLI telemetry. No SaaS. No remote LLM calls.
projectmem is a local-first memory layer for AI coding agents (MCP server) — it captures issues, attempts, fixes, and decisions, then warns at git commit before you repeat a mistake. Python, runs locally, works with Claude, Cursor, Antigravity, and Codex
The most comprehensive MCP server for VS Code — 100+ tools across 25 categories. File ops, terminal, git, LSP providers (hover, completion, definition, references), and more. Free core + premium features.
Official GitHub MCP server for repository search, issues, pull requests, code context, and GitHub workflows in MCP-compatible AI clients.
Creates commit messages from staged files in a local git repository.
An MCP server for interacting with the AtomGit API for version control and code hosting.
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
