CLI Reference
The plugin’s SessionStart hook provisions the cc-review binary pinned by the plugin version — a brew install wins, otherwise a checksum-verified download to the plugin data dir — and points the plugin/bin/cc-review symlink at it; that directory is on PATH in plugin sessions. Every command is a thin shell around the daemon control client; the daemon lazy-starts on first use.
start
cc-review start [--session <id>] [--cwd <dir>] [--new] [--base <ref>] [--pr <ref>] [--open]Start or resume a review of the working tree. Prints, in order: the review URL; zero or more tailnet: <url> lines, the same review served on the daemon’s tailnet addresses, printed when synckit’s mesh trust is on, as Internals describes; channel: active|pending|inactive — active when this window’s channel is proven (its first delivered tag was acknowledged via channel-ack) and the channel consumer is attached, pending when the channel server is attached but the window is unproven — the normal state before any tag has been acked; that start also injects a one-shot channel.probe tag to solicit the proof — inactive when no channel consumer exists; setup: {"offer":…,"reason":…} — the first-run channel-approval offer (always printed; an offer-check error degrades to offer: false with the error as the reason); and organize: <AI request JSON> — the daemon’s eager organize request, printed when a new version was created and re-offered (same id) when an unchanged resume finds the latest version still unorganized. In a Graphite-tracked repo a stack: {"trunk":…,"branches":[…]} line also prints: the review is a stack review — one section per stacked branch diffed against its parent, plus a pending working-tree section — re-resolved on every snapshot with nothing pinned. With --pr, the review is a PR review instead: one section per pull request in the stack, loaded from GitHub. It prints a pr: <owner>/<name>#<N> (stack: #<a> #<b> …) line instead of stack:. When Claude’s GitHub replies can’t post yet, the setup: JSON gains a github key holding what’s missing: cc-review github setup, the App’s install URL, or the check’s error. See Reviewing pull requests.
| Flag | Type | Default | Description |
|---|---|---|---|
--session |
string | "" |
Claude session id (keys the review with the repo root) |
--cwd |
string | "" |
working directory (defaults to the current directory) |
--new |
bool | false |
force a fresh review, detaching any existing one for this session |
--base |
string | "" |
pin a new review’s diff base: the fork point of this ref and the working copy (default: HEAD, falling back to trunk when the working tree is clean); in a Graphite-tracked repo, forces a flat single-diff review instead of the auto-detected stack review |
--pr |
string | "" |
review a GitHub pull request and its stack: a PR URL, owner/name#N, #N, or N (bare forms use the cwd’s origin remote); mutually exclusive with --base |
--open |
bool | false |
open the review in the browser after printing its URL |
cc-review start --session "$CLAUDE_SESSION_ID"Review pull request 123 of the cwd’s repo and open it in the browser:
cc-review start --session "$CLAUDE_SESSION_ID" --pr '#123' --openwatch
cc-review watch [--session <id>] [--cwd <dir>]Stream review events as line-delimited JSON, one event per line, then exit on the terminal submit event. Output is line-buffered and resumes from a persisted cursor, so a restart re-delivers nothing it already emitted. Meant to run under a Claude Code Monitor so each line becomes a chat notification.
| Flag | Type | Default | Description |
|---|---|---|---|
--session |
string | "" |
Claude session id |
--cwd |
string | "" |
working directory (defaults to the current directory) |
cc-review watch --session "$CLAUDE_SESSION_ID"reply
cc-review reply --comment <id> [--kind <kind>] [--body <text>] ...
cc-review reply --answer-to <id> (--answer <text> | --select <label> [--other <text>] [--notes <text>])Post a Claude question, ask, or clarification under a comment, or answer one. Returns immediately. The two modes are exclusive. --comment posts a new reply; --answer-to records an answer to an existing one. Flags from the other mode are rejected, not ignored.
| Flag | Type | Default | Description |
|---|---|---|---|
--comment |
int64 | 0 |
comment id to reply under |
--kind |
string | clarification |
reply kind: question | ask | clarification |
--body |
string | "" |
reply text (the question for --kind ask) |
--header |
string | "" |
short chip shown on the ask card, e.g. Approach |
--multi-select |
bool | false |
allow picking multiple options |
--options-json |
string | "" |
JSON array of {"label","description"?,"preview"?} (requires --kind ask) |
--answer-to |
int64 | 0 |
the reply id of the question or ask being answered |
--answer |
string | "" |
the answer text when answering a plain question |
--select |
string array | [] |
a chosen option label when answering an ask (repeatable) |
--other |
string | "" |
free-text answer outside the offered options |
--notes |
string | "" |
a note riding along with the selection |
Post an ask card under comment 12:
cc-review reply --comment 12 --kind ask \
--body "How should the limit be configured?" --header "Config" \
--options-json '[{"label":"Env var"},{"label":"Per-route option"}]'Answer that ask:
cc-review reply --answer-to 7 --select "Env var" --notes "Default to 100"feedback
cc-review feedback [--session <id>] [--cwd <dir>]Print the frozen feedback JSON after Submit. The payload contains the comment threads and any open questions.
| Flag | Type | Default | Description |
|---|---|---|---|
--session |
string | "" |
Claude session id |
--cwd |
string | "" |
working directory (defaults to the current directory) |
cc-review feedback --session "$CLAUDE_SESSION_ID"status
cc-review status [--session <id>] [--cwd <dir>]Show daemon and review status. Prints the daemon version, HTTP port, and the review id and state for this session/repo. Reports daemon: not running without spawning one.
| Flag | Type | Default | Description |
|---|---|---|---|
--session |
string | "" |
Claude session id |
--cwd |
string | "" |
working directory (defaults to the current directory) |
cc-review statuslist
cc-review list [--cwd <dir>]List every open or expired review across repos, one SLUG STATUS AGE IDLE SCOPE row each: status, age since creation, idle time since the last real activity, and the repo. Real activity is comments, replies, AI-bar requests, and new versions; presence pings don’t count. Prints no open reviews when there are none. The listing spans all repos and runs from anywhere.
| Flag | Type | Default | Description |
|---|---|---|---|
--cwd |
string | "" |
working directory (defaults to the current directory) |
cc-review listclose
cc-review close [review] [--session <id>] [--cwd <dir>] [--stale]Close a review without submitting. With no argument, closes the current window’s review; pass a slug or id to close any review. --stale instead closes every expired review across all repos — sweeping open reviews idle past the 24-hour TTL into expired first — and rejects an explicit review argument. A closed review never resumes; an expired one stays bound to its window, so an explicit start reopens it. Prints one line per review closed. --stale spans all repos and runs from anywhere.
| Flag | Type | Default | Description |
|---|---|---|---|
--session |
string | "" |
Claude session id (keys the review with the repo root) |
--cwd |
string | "" |
working directory (defaults to the current directory) |
--stale |
bool | false |
close every expired review across all repos, sweeping idle ones past the TTL first |
cc-review closestop
cc-review stopStop the background daemon. No flags.
cc-review stopsetup-channels
cc-review setup-channels [--check | --apply | --decline]Make cc-review an approved Claude channel, silencing the dev-channels warning. Hidden from --help. cc-review start already prints the offer check as its setup: line; /cc-review:start reads that line and runs --apply or --decline once based on the user’s answer. The three flags are mutually exclusive; passing none runs the check.
| Flag | Type | Default | Description |
|---|---|---|---|
--check |
bool | false |
print {offer,reason} JSON for the first-run offer (default; the same JSON start prints as setup:) |
--apply |
bool | false |
write the approved-channels config (prompts for admin) |
--decline |
bool | false |
record that the offer was declined |
cc-review setup-channels --checkgithub
cc-review github setup [--org <org>]
cc-review github status [--cwd <dir>]Create and inspect the cc-review GitHub App that Claude’s PR-review replies post as. setup prints open: <url>, plus a tailnet: <url> line per tailnet address the daemon serves, and opens the daemon’s setup page, which hands GitHub an app manifest. Open a tailnet: URL instead when your browser is on another machine; GitHub sends you back to whichever address you opened. After you confirm on GitHub, the browser lands on a daemon page that offers the App’s logo for download, links to the App’s settings page where you upload it, and then links to the App’s install page. setup prints app: <slug>, logo: <path> (the same PNG, written to ~/.cc-review/v1/github-app-logo.png), and install: <url>. It waits up to 10 minutes for you to finish. status prints app: <slug>, bot: <login>, and install: installed on <owner>/<name> or install: not installed on <owner>/<name>; install at <url> for the cwd’s repo, or app: not set up; run `cc-review github setup` when there’s no App yet.
| Flag | Type | Default | Description |
|---|---|---|---|
--org |
string | "" |
create the app under this GitHub organization instead of your account (setup only) |
--cwd |
string | "" |
working directory (defaults to the current directory; status only) |
cc-review github setup
cc-review github status