av-cli（AV CLI 命令行）

Aviator CLI (av) for Stacked PRs

You are helping the user work with stacked pull requests using Aviator's av CLI tool.

IMPORTANT: NEVER modify the av/av.db file inside the git dir (found via git rev-parse --git-common-dir) directly. This JSON file is managed by av commands. You may read it to understand stack structure, but always use av CLI commands to make changes.

First Steps (ALWAYS DO THIS)

When working with av, ALWAYS read av/av.db from the git common dir first to understand the branch structure. This JSON file is the source of truth.

git rev-parse --git-common-dir

Then use the result to read the db file:

cat <git-common-dir>/av/av.db

IMPORTANT: Always use cat to read av.db, NEVER the Read tool. The .db extension causes Read to incorrectly treat it as binary.

Tip: Use git branch --show-current to identify the current branch, then look up that branch's entry in av.db for its stack context and PR info.

Do NOT use av tree to understand structure - its visual output is misleading. Only use av.db.

Critical Rules

NEVER use git commit or git push directly. Always use av commit for commits and av pr (which pushes automatically) or av sync --push=yes --prune=yes for pushing. Using git directly skips restacking and breaks the stack.

NEVER pass --no-edit to av commit --amend. The flag doesn't exist — no-edit is already the default behavior. Just use av commit --amend.

Detection & Setup

Check if av is initialized: run git rev-parse --git-common-dir, then test -f <git-common-dir>/av/av.db

• If the file exists, the repo is av-initialized. Use av commands for branch/PR operations.
• If the file does NOT exist, ask the user if they want to initialize with av init.
• For detailed command reference, see reference.md.
• For workflow examples, see examples.md.
• Run av <command> --help or man av-<command> for up-to-date command documentation.

Non-Interactive Mode (Agents & Automation)

Many av commands default to interactive TUI prompts that agents cannot use. Always use the non-interactive flags listed below.

Critical syntax note: Flag values require = (equals sign), not a space. --push=yes works; --push yes does NOT.

Recommended agent workflow:

# Creating a new PR:
av commit -A -m "message"
av pr --title "Title" --body "Body"  # pushes the branch and creates the PR

# Pushing updates to an existing PR (single branch, not in a stack):
av commit -A -m "message"
av pr  # no args needed — just pushes the branch and updates the PR, no editor prompt

# Pushing updates when working in a stack (syncs the entire stack):
av commit -A -m "message"
av sync --push=yes --prune=yes

Understanding Stack Structure

Run git rev-parse --git-common-dir, then cat <result>/av/av.db to understand branch relationships. Format:

{
  "branches": {
    "feature-api": {
      "name": "feature-api",
      "parent": { "name": "master", "trunk": true },
      "pullRequest": {
        "number": 123,
        "permalink": "https://github.com/org/repo/pull/123"  // USE THIS URL
      }
    },
    "feature-ui": {
      "name": "feature-ui",
      "parent": { "name": "feature-api", "head": "abc123" }
    }
  }
}

Key fields:

• parent.trunk: true → branch is directly off main/master (not stacked)
• parent.name + parent.head → branch is stacked on another branch
• pullRequest.permalink → full PR URL (always use this when displaying PR info, not just the number)
• excludeFromSyncAll: true → branch excluded from av sync --all

Reading the structure: Each branch's parent.name tells you what it's based on. Build the tree by following parent relationships. Branches with trunk: true are all independent roots.

Core Concepts

What is a Stack?

A stack is a chain of dependent branches where each branch builds on the previous one:

main
 └── feature-auth (PR #1: adds auth)
      └── feature-login (PR #2: adds login, depends on auth)
           └── feature-logout (PR #3: adds logout, depends on login)

Why Use av?

1. Automatic rebasing: When you update a branch mid-stack, av rebases all child branches.
2. Correct PR bases: av pr automatically sets the correct base branch (parent branch, not trunk).
3. Coordinated syncing: av sync fetches, rebases, and pushes all branches in one command.

How Stacked PRs Work

Each PR shows only its own diff (changes relative to its parent branch), not the cumulative diff against trunk. This makes code review focused and manageable.

Common Use Case: Full-Stack Features

The most common workflow is building features that span multiple layers:

main
 └── add-feature-db       (DB schema/migrations)
      └── add-feature-service  (Backend service logic)
           └── add-feature-api      (REST/GraphQL endpoints)
                └── add-feature-ui       (Frontend components)

Each layer gets its own focused PR. Reviewers with different expertise (DBA, backend, frontend) review the relevant parts. When the DB schema changes during review, av sync propagates updates through the entire stack automatically.

Essential Commands

Branch Management

av adopt --remote is particularly useful for pulling in a colleague's branch (or any remote branch not yet tracked locally) and automatically setting up the correct stack relationships. It fetches the branch, detects its parent chain, and integrates it into your local av state — much more reliable than manually checking out and adopting.

Committing

av commit vs git commit — key differences:

• --amend reuses the existing message by default. There is no --no-edit flag — it is the default behavior.
• --edit explicitly opens the editor when amending (the opposite default from git).
• -a behaves the same as git's -a (stages modified/deleted tracked files only).
• -A / --all-changes is av-specific: stages ALL files including untracked (git has no equivalent single flag).

Selective staging: -a and -A stage everything (tracked or all). To commit only specific files, stage them first with git add <files>, then run av commit without -a/-A. This works for both new commits and amends.

Common mistakes:

• git commit -m "message" → use av commit -m "message" instead
• git push → use av pr when creating a PR (it pushes automatically); use av sync --push=yes --prune=yes to push changes and rebase across the stack
• av pr --title "..." --body "..." when PR already exists → just av pr (no args needed, no editor prompt)
• av sync --push=yes --prune=yes just to push a single standalone branch → use av pr instead (but av sync is correct when working in a stack)
• av commit --amend --no-edit → just av commit --amend (no-edit is the default, the flag doesn't exist)
• av sync before/after av pr → unnecessary; av pr pushes on its own

Pull Requests

av pr automatically pushes the current branch to the remote before creating or updating the PR. No prior av sync or git push is needed.

Note: Pass --title and --body when creating a new PR to avoid editor prompts. When the branch already has a PR, bare av pr just pushes — no editor, no prompt.

Synchronization

Non-interactive mode: av sync prompts for confirmation by default. Use explicit flags (note the = syntax — a space does not work):

av sync --push=yes --prune=yes          # Sync current stack
av sync --all --push=no --prune=yes     # Sync all stacks (use after PRs are merged, ask user before pushing)

Both --push and --prune must always be specified — omitting either triggers a TUI prompt. Options: yes, no, or ask (default). --prune=yes is safe in most cases — it only deletes local branches whose PRs have already been merged. Use --prune=no if you need to keep merged branches around locally (e.g., for reference or if other worktrees have them checked out).

Timeout: av sync performs a fetch + rebase + push cycle and can take 15-30+ seconds. Use a timeout of at least 60 seconds for any av sync command.

Navigation

Conflict Resolution

When rebasing causes conflicts:

1. Resolve conflicts in your editor
2. Stage resolved files with git add
3. Continue with av sync --continue or av restack --continue

Or abort with --abort, or skip the problematic commit with --skip.

Using av for All Workflows

Once a repo is av-initialized, use av for everything - even single PRs. av works great for non-stacked workflows too and keeps things consistent.

av pr vs av sync: av pr pushes the current branch and creates/updates its PR — it's the simplest way to push a single standalone branch. av sync fetches, rebases, and pushes across the entire stack — use it when working in a stack so all branches stay in sync, and for cleanup after merges.

Important Behaviors

1. Use av commands consistently - they work for single PRs and stacks alike.
2. av commit auto-restacks child branches when you have them.
3. Let av pr set the base - don't manually specify base branches.
4. After PR merges, run av sync --all --push=no --prune=yes to clean up and rebase remaining branches. Ask the user before pushing all stacks.
5. Don't mention stacks in commits/PRs - never reference stack position, parent branches, or stack relationships in commit messages, PR titles, or PR bodies. The av tooling handles this metadata automatically.
6. Always show full PR URLs - when displaying PR info, use the permalink field from av.db. Never show just "PR #123" - always show the full clickable URL like https://github.com/org/repo/pull/123.