Skill

---
name: chip
description: Use when the user runs /chip or asks which PRs are waiting for their review (in English or Vietnamese, e.g. "which PRs need my review", "review inbox") and wants to pick some to review with their configured review skill.
---

# chip — PR review inbox

`{{CHIP}}` does all the data work. You only pass its JSON to AskUserQuestion, collect the picks, and run the review loop. Never build, re-sort or filter options yourself, never reword them, and never print the PR list as a table.

The main way to pick PRs is **in the prompt bar**: the `chip` MCP server exposes every waiting PR as a resource, so typing `@` (or `@chip:`) followed by a few letters of the repo, number or title shows them in the autocomplete. The user can mention several.

Args:
- **Args contain `pr://`** (e.g. `/chip @chip:pr://shopbox-api/274-fix-orders… @chip:pr://api/2069-…`): run `{{CHIP}} pick "<args verbatim>"` and go straight to step 4 (review loop) with its output. Show no menus. If it exits 2, show stderr and tell the user the list may be stale; they can ask to reload, which runs `chip repos --refresh --force`, and mention again.
- **No args**: reply with exactly one line, "Type `/chip @` then a repo name, PR number or keyword to pick PRs (several allowed), and press Enter to review. For menus: `/chip menu`." and stop.
- `/chip menu [words]`: the in-chat menus below (steps 1–3). With words, skip step 1 and start step 2 with `--q "<words>"`.
- `/chip terminal`: run `{{CHIP}} open`, which opens the fzf picker in a new Terminal window. Reply with its stdout (or stderr on failure) and stop.

## 1. Choose the project (one keystroke)

```bash
{{CHIP}} repos --refresh     # fetches from GitHub (~5s), or reuses a fetch < 3 min old; add --force if the user asks to reload
                                         # later in the flow: `chip repos` (no --refresh) reuses the same fetch
```

Output: `{total, hidden: {approved, draft}, updated, questions}`. `questions` holds ONE single-select question: `All` plus the 3 busiest repos. The other repos are listed in the question text for typing.

- Exit 1: show stderr as-is (it includes the `! gh auth login` hint when relevant) and stop.
- `total == 0`: "Inbox zero 🎉" plus the hidden counts, then stop.
- Otherwise, print one line, `<total> PRs waiting for you (hidden: <approved> approved, <draft> drafts · updated <updated>)`. Then call **AskUserQuestion with `questions` exactly as given**.
- Answer `All`: no filter. A repo label: `--repo <label>`. Other text: if it equals or contains a repo name from the question, use `--repo <that name>`; otherwise use `--q "<text>"`.

## 2. Tick PRs

```bash
{{CHIP}} menu --cached [--repo a] [--q "<words>"] [--page N]
```

(For `/chip menu <words>`, where step 1 was skipped, the first call is `chip menu --q "<words>"` without `--cached`. It fetches from GitHub.)

The output is `{page, pages, total, filter, hidden, questions}`. Call **AskUserQuestion with `questions` exactly as given**. If `total == 0`, say "No PRs match `<filter>`" and go back to step 1.

When the result has **16 PRs or fewer**, `questions` contains only multi-select PR questions and no nav question: the user ticks and submits once. Collect the ticked labels, ignoring `None`, and go straight to step 3. If an Other answer holds text instead, treat that text like the Other answer of step 1 (switch project or search), keep the labels ticked so far, and rerun this step.

When the result has **more than 16 PRs**, there are pages of 12 plus a nav question "Next?". Keep the filter state and **every label picked so far** across pages:

| Nav answer | Do |
|---|---|
| `Next page` | the same command with `--page <page+1>` |
| `Change project` | step 1 again, using `chip repos` without `--refresh` |
| `Search` | ask in plain text "Type keywords (title, author, Jira, repo) — or `clear`:" and wait, then rerun with `--q "<text>"` |
| Other (free text) | the same as the Other answer of step 1 |
| `Review selected PRs` | go to step 3 |

If nothing was picked at review time, say so in one line and stop.

## 3. Pick

```bash
{{CHIP}} pick "<label1>,<label2>,…"
```

On exit 2, show the error and re-ask. On exit 0 the output is a JSON array of `{index, label, repo, number, title, url}` in list order.

## 4. Review loop — one PR at a time

For item i of n:

1. Print `[i/n] <label> <title>`.
2. Run `{{CHIP}} repo <repo>` to get the local path.
   - On exit 2, ask with AskUserQuestion "No local clone of <repo>. Clone it into ~/.cache/chip/repos?" (Clone / Skip this PR). On Clone, run `{{CHIP}} repo <repo> --clone`. If the user chooses Skip or the clone fails, tell the user and skip to the next PR.
3. Run `{{CHIP}} prompt <label>` to get the review prompt from the user's config (the first skill). If it starts with `/<name> `, invoke the skill `<name>` with the rest as args, followed by ` — local clone: <path>. The shell cwd resets between commands, so run every git/gh command as \`cd <path> && …\`.` Otherwise, follow the prompt text as the instruction, working in `<path>`.
4. After the review report, if there is a next PR, ask with AskUserQuestion "Continue with <label>?" (Continue / Stop) and stop on Stop.

At the end, list which PRs were reviewed and which were skipped.

## Rules

- Never review a PR the user did not pick.
- Do not run reviews in parallel.
- Do not `git checkout`/`stash` in the user's clones; the review skill already reads refs with `git show`.