| name | meebox |
| description | Review and act on pull requests through the Code Meeseeks "meebox" CLI — list and inspect PRs, run the AI review agent, and record review outcomes (approve / needs-work / comment). Use when the user wants to triage, review, or act on pull requests via a running Code Meeseeks desktop app. |
meebox — pull-request review over the local API
meebox is a thin cross-platform CLI over a running Code Meeseeks desktop app's
local HTTP API. Use it to browse pull requests, drive the review agent, and record
review outcomes. Output defaults to YAML; pass --output json when parsing results.
Prerequisites
- The Code Meeseeks desktop app is running with the local API service enabled
(Settings → Integration).
- The
meebox binary is available — it ships in this skill directory; put it on PATH
or invoke it by path (./meebox).
- Connection is configured (below). Verify with
meebox whoami.
Connect
Provide the API base URL + token explicitly — the CLI never reads the app's config.yaml.
Easiest is meebox login (persists to ~/.code-meeseeks/cli.yaml); or use env vars:
meebox login --token <token>
meebox login --token <token> --server http://host:18765
export MEEBOX_API_URL=http://127.0.0.1:18765
export MEEBOX_TOKEN=<token>
meebox whoami
meebox --output json pr list | jq '.[].id'
Command map
Root-level meebox whoami and meebox version need no PR. The rest split into two domains —
pr and agent; PR-scoped subcommands take the required --pr <id> flag (id from pr list),
while pr categories / pr refresh / pr list are collection-level (no --pr).
Browse / inspect — pr
meebox pr categories — the active platform's categories / statuses filter vocabulary for pr list.
meebox pr refresh — trigger one immediate poll for the latest PRs (same as the app's manual refresh); returns change counts (fetched / changed / added / removed / errors). Run before pr list for fresh data.
meebox pr list [--category review-requested|created|assigned|mentioned] [--status pending|approved|needs_work|conflict|mergeable] [--query <text>] [--skip N] [--limit N] — paginated (default limit 100), slim fields (id / title / author / createdAt first).
meebox pr show --pr <id> — full detail incl. description.
meebox pr diff --pr <id> [--file <path> --side base|head] — changed files, or one file's content.
meebox pr activity --pr <id> · meebox pr commits --pr <id> · meebox pr reviewers --pr <id>.
Review agent — agent
meebox agent review --pr <id> — run the auto-review micro-flow (describe→review→[ask]→summary).
meebox agent status --pr <id> · meebox agent history --pr <id> — progress + conversation.
meebox agent instruct --pr <id> <describe|review|ask|improve> [text] — one read-only tool call.
meebox agent chat --pr <id> <message> — natural-language message (may trigger tasks).
meebox agent run list --pr <id> · meebox agent run cancel --pr <id> --run <runId> — inspect / cancel a single agent run.
meebox agent stop --pr <id> — stop the whole agent for the PR.
Record outcomes — pr (real remote writes)
meebox pr approve --pr <id> · meebox pr needswork --pr <id> — post a review decision.
meebox pr comment --pr <id> <message> — post a top-level comment.
Root-level (no PR):
meebox login --token <token> [--server <url>] — save credentials to cli.yaml (default server is loopback); later commands need no flags/env.
meebox whoami — current user + platform + connection (confirm your token resolves).
meebox version — CLI (client) + app (server) versions; client-only when the server is unreachable.
meebox skill — print this doc (SKILL.md is embedded in the binary), so usage is retrievable even without the file.
Typical loop
meebox --output json pr list --status pending → choose a PR id.
meebox pr show --pr <id> and/or meebox agent review --pr <id>; poll meebox agent status --pr <id>.
- Inspect changes:
meebox pr diff --pr <id>.
- Record the outcome:
meebox pr approve --pr <id> / meebox pr needswork --pr <id> / meebox pr comment --pr <id> "…".
Boundaries
- Not available: merging PRs, and the agent's publish / mutating tools —
instruct accepts
only the read-only tools describe / review / ask / improve.
- Exit codes:
0 ok · 2 auth failure · 3 not found · 1 other; errors print to stderr with the API error code.