gerrit

# Gerrit Invoke with **`@gerrit`** or **`/gerrit`**, then state what you want — e.g. `inbox`, `diff 46568`, `what CLI commands are available?`, `review 46568 +1`. Gerrit exposes a **remote CLI over SSH**: `ssh -p <port> <user>@<host> gerrit <subcommand>`. This skill runs those commands from the **current git repo** (must be a Gerrit project). Connection details come from [`.gitreview`](query-reference.md#env) or `git remote`; the live `gerrit` / `gerrit <cmd> --help` output is the source of truth for subcommands and flags. --- ## Workflow ### Step 1 — Resolve environment Read [`query-reference.md`](query-reference.md#env) and resolve **host**, **port**, **user**, **project** from `.gitreview` or `git remote get-url origin`. Probe with: ```bash ssh -o ConnectTimeout=10 -o BatchMode=yes -p <port> <user>@<host> gerrit version ``` **1b. `jq` (when this turn parses `--format=JSON`)** ```bash command -v jq ``` - **query**, **diff** metadata, or **review** without an `N,PS` target → present: continue; missing: note absence, inline parsing is acceptable, but do not create a `.py` file for JSON - **cli** explore or no JSON parsing → skip 1b **Done when:** host, port, user, and project are known (or SSH failure stops execution), and jq status is available / missing-with-fallback-noted / skipped. ### Step 2 — Classify branch Pick exactly one branch from the user's message (after `@gerrit` / `/gerrit`): | Branch | When | |--------|------| | **query** | inbox, dashboard, open changes, pending reviews, or a Gerrit search string | | **diff** | change number + read or review the patch | | **review** | score, comment, LGTM, or submit | | **cli** | explore available subcommands, `--help` for a command, or run any other `gerrit` subcommand | **Routing defaults:** change number + read code → **diff**; change number + score/submit → **review**; inbox / my changes / pending reviews → **query** (inbox presets); unfamiliar subcommand name → **cli**. **Done when:** exactly one branch is selected and parameters (change number, query string, labels, subcommand) are known or extracted from the user message. ### Step 3 — Execute branch - **query** — [`query-reference.md`](query-reference.md): inbox presets (three fixed queries) or custom `gerrit query --format=JSON …`; pipe output through `jq` - **diff** — [`diff-reference.md`](diff-reference.md): metadata query → `git fetch` change ref → `git diff` - **review** — [`review-reference.md`](review-reference.md): resolve `CHANGE,PATCHSET` + `--project` → dry-run → execute - **cli** — [`cli-reference.md`](cli-reference.md): **explore** (`gerrit` list + `gerrit <cmd> --help`) or **execute** (help first, then run; mutating commands dry-run) **Done when:** branch output is complete (table, diff, dry-run proposal, or command output), or failure is explained with the SSH/git error. ### Step 4 — Present Deliver a concise summary: - Resolved **host**, **project**, and branch used - Results (table, diff excerpt, or command output) - Change URLs: `https://<host>/c/<project>/+/<change-number>` - For **cli** explore mode: note that naming a subcommand and args will run or dry-run it **Done when:** results are in the chat. --- ## Boundaries - Search, inbox, custom `gerrit query` → **query** - Fetch patch / AI code review on a change → **diff** - `gerrit review` scores and submit → **review** - List subcommands, `--help`, `set-reviewers`, `ls-projects`, `stream-events`, admin commands → **cli** - GitHub PR workflows (`gh pr`) → out of scope - Gerrit REST API over HTTPS (often needs SSO) → out of scope unless user insists; prefer SSH --- ## Reference — key constraints - `gerrit query --files` returns file names and line counts only — **not** diff text; use **diff** branch for patches. - `gerrit query` date filter: `before:` is **exclusive** (day after the inclusive end date). - Never hardcode host, user, or project — resolve per run from the repo environment. - **review** branch: `CHANGE,PATCHSET` target and `--project <project>` before SSH — see [`review-reference.md`](review-reference.md). - **jq first** — `--format=JSON` → pipe through `jq`; if `jq` is missing, inline parsing is acceptable; do not create temporary `.py` files for JSON. - **NDJSON** — one object per line; filter results with `select(.type != "stats")`. - **jq hygiene** — scalars: `jq -r .field`; long text: `.subject | .[0:70]`; array fields: `// []` fallback.