Skip to content
git push no-mistakes

Quick Start

This walks you through your first gated push. For install options other than the macOS/Linux one-liner, see Installation.

1. Install

Terminal window
curl -fsSL https://raw.githubusercontent.com/kunchenguid/no-mistakes/main/docs/install.sh | sh

The installer drops the binary in ~/.no-mistakes/bin, links it into ~/.local/bin or /usr/local/bin, and restarts the background daemon. If the restart fails, the install command fails.

Official release binaries installed this way include the default self-hosted telemetry host and website ID. Disable telemetry with NO_MISTAKES_TELEMETRY=0, or override the host and website ID with NO_MISTAKES_UMAMI_HOST and NO_MISTAKES_UMAMI_WEBSITE_ID.

2. Check prerequisites

Terminal window
no-mistakes doctor

You need:

  • git
  • One supported agent binary (claude, codex, acli for Rovo Dev, opencode, or pi), or a separately installed acpx binary for agent: acp:<target>
  • For PRs and CI: gh (GitHub), glab (GitLab), or Bitbucket Cloud credentials

For ACP agents, verify acpx or acpx_path separately because no-mistakes doctor does not validate ACP targets.

See Provider Integration for PR/CI setup.

3. Initialize a repo

Navigate to any git repo with an origin remote:

Terminal window
no-mistakes init

This creates or refreshes a local bare repo at ~/.no-mistakes/repos/<id>.git, installs a post-receive hook, best-effort isolates the gate’s hooks path from shared local Git config writes when Git supports config --worktree, adds or repairs a no-mistakes git remote in your working repo, installs the /no-mistakes agent skill, and ensures the daemon is running.

$ no-mistakes init
  ✓ Gate initialized


    repo  /Users/you/src/my-repo
    gate  no-mistakes → /Users/you/.no-mistakes/repos/abc123def456.git
  remote  git@github.com:you/my-repo.git
   skill  /no-mistakes installed for agents at user level


  Push through the gate with:
  git push no-mistakes <branch>

origin is unchanged. If you need to bypass the gate for a specific push, use git push origin <branch>.

You can safely re-run no-mistakes init later to refresh gate wiring or update the installed agent skill after an upgrade. If you rename or move the repo directory, re-run no-mistakes init from the new path to reattach the existing gate and keep its run history. Copied repos get their own fresh gate while the original path still exists.

4. Push through the gate

Instead of git push origin, push to the no-mistakes remote:

Terminal window
git checkout -b feature/login-fix
# do work, commit...
git push no-mistakes

The push lands in the local bare repo, the hook notifies the daemon, and the daemon starts the pipeline in a disposable worktree.

5. Watch the pipeline

Terminal window
no-mistakes

If the current branch has an active run, this attaches directly. If not, the setup wizard can walk you through creating a branch, committing, and pushing through the gate, then attach if the daemon registers the new run. By default that path is interactive in a TTY. With no-mistakes -y, the wizard accepts defaults automatically, stays visible and auto-advances in a TTY, and falls back to the headless path without a TTY.

The TUI shows each step’s progress, streams agent output, and pauses for your approval when findings need attention. See Using the TUI for keybindings and layout.

Or let your agent run the gate

If you are already working inside a coding agent like Claude Code, you don’t have to switch to the terminal at all. no-mistakes init installed the /no-mistakes skill at user level, available in every repo, so you can ask the agent to do a task and gate it:

/no-mistakes add a --json flag to the status command

Or, if the work is already committed on a feature branch, use bare /no-mistakes to validate it:

/no-mistakes

In task-first mode, the agent inspects scope, preserves unrelated work, commits only the task changes on a feature branch, and passes your task text as --intent. In validate-only mode, it validates the existing committed work. Either way, it applies low-risk fixes itself and stops to relay any finding that needs your judgment. It drives the same gate as the TUI through no-mistakes axi, a non-interactive command surface that uses flags only, prints TOON on stdout, and exposes the same approval gates through no-mistakes axi respond.

See Driving no-mistakes as an agent for the full agent workflow.

What happens next

The pipeline runs these steps in order:

  1. Intent - use agent-supplied intent when present, otherwise infer author intent from recent local agent transcripts
  2. Rebase - onto the latest upstream
  3. Review - AI code review of your diff
  4. Test - baseline tests plus evidence checks when intent is known
  5. Document - updates docs and reports unresolved gaps
  6. Lint - your linters (configured command or agent-detected)
  7. Push - to the real upstream remote
  8. PR - create or update the pull request
  9. CI - poll CI, watch PR mergeability, auto-fix failures

Steps that find issues pause for your approval. See the Pipeline concept page for the overview and Pipeline Steps for each step’s exact behavior.