oliver-kriska/claude-elixir-phoenix / plugins/elixir-phoenix/skills/pr-review
pr-review
VerifiedAddress feedback left on a GitHub pull request: fetch unresolved review threads, make agreed Elixir/Phoenix code fixes, reply, and resolve. Use for a PR URL/number or reviewer comments. NOT for pre-PR review, findings triage, or CI monitoring.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.0 KB |
| references/bot-triage.md | markdown | 3.5 KB |
| references/gh-commands.md | markdown | 4.4 KB |
| references/response-patterns.md | markdown | 4.3 KB |
references/gh-commands.md
View on GitHub# gh Command Reference — PR Review Threads
Verified against gh 2.94.0 and the live GitHub API (2026-06). Repo is
inferred from cwd, or parsed from a PR URL, or passed via `-R owner/repo`.
## The three comment surfaces
These are different endpoints with different reply mechanisms. Mixing them
up means replies silently land in the wrong place.
| Surface | Read | Resolvable? | Reply |
|---------|------|-------------|-------|
| Inline review comment (file/line) | GraphQL `reviewThreads` or REST `pulls/{n}/comments` | YES (thread) | REST `.../comments/{id}/replies` or GraphQL thread reply |
| Review summary ("Changes requested" body) | REST `pulls/{n}/reviews` | NO | Top-level conversation comment, or address inline and re-request review |
| Top-level conversation comment | REST `issues/{n}/comments` (a PR IS an issue) | NO | `POST repos/{o}/{r}/issues/{n}/comments -f body=...` |
## PR metadata
```bash
gh pr view "$PR" --json number,title,state,baseRefName,headRefName,url,author \
--jq '{number,title,state,base:.baseRefName,head:.headRefName,url,author:.author.login}'
```
`gh pr view` accepts a number OR a URL and resolves the repo from the URL.
## Review threads with resolve state (GraphQL — the core query)
REST review comments have NO `isResolved` field and NO thread node ID.
The GraphQL query in SKILL.md Step 1 is the only way to get both. Key fields:
- `id` — `PRRT_...` thread node ID → used by resolve/unresolve and GraphQL reply
- `comments.nodes[].databaseId` — integer REST comment ID → used by REST reply
- `line` may be `null` on outdated threads → fall back to `originalLine`
- `--paginate` auto-follows `endCursor` because the query exposes
`pageInfo { hasNextPage endCursor }` and accepts `$cursor`
- With `--paginate`, `--jq` runs PER PAGE. Add `--slurp` to merge pages into
one array when counting/sorting across pages
## Reply
REST (preferred when `databaseId` is in hand — reply to the thread's ROOT comment):
```bash
gh api --method POST \
"repos/$OWNER/$REPO/pulls/$PR/comments/$COMMENT_ID/replies" \
-f body="$REPLY_TEXT" --jq '{reply_id: .id, in_reply_to: .in_reply_to_id}'
```
GraphQL (when only the thread ID is in hand):
```bash
gh api graphql \
-f query='mutation($threadId:ID!,$body:String!){
addPullRequestReviewThreadReply(input:{pullRequestReviewThreadId:$threadId, body:$body}){
comment { id databaseId url } }}' \
-F threadId="$THREAD_ID" -F body="$REPLY_TEXT"
```
## Resolve / unresolve
```bash
gh api graphql -f query='mutation($threadId:ID!){
resolveReviewThread(input:{threadId:$threadId}){ thread { id isResolved } }}' \
-F threadId="$THREAD_ID" --jq '.data.resolveReviewThread.thread'
gh api graphql -f query='mutation($threadId:ID!){
unresolveReviewThread(input:{threadId:$threadId}){ thread { id isResolved } }}' \
-F threadId="$THREAD_ID"
```
An invalid node ID returns a clean `NOT_FOUND` — safe to surface as an error.
## Review summaries + conversation comments
```bash
# Review summaries (incl. bot passes — Copilot/Codex post these)
gh api "repos/$OWNER/$REPO/pulls/$PR/reviews" \
--jq '.[] | {reviewer: .user.login, isBot: (.user.type=="Bot"), state, body}'
# Top-level conversation: read + post (post = outward-facing, confirm first)
gh api "repos/$OWNER/$REPO/issues/$PR/comments" --jq '.[] | {user: .user.login, body}'
gh api --method POST "repos/$OWNER/$REPO/issues/$PR/comments" -f body="$SUMMARY"
```
## Bot detection
| Source | Field | Bot value |
|--------|-------|-----------|
| GraphQL | `author.__typename` | `"Bot"` |
| REST | `user.type` | `"Bot"` |
| REST | `user.login` suffix | `[bot]` — **NOT reliable**: `pulls/{n}/comments` reports `Copilot` (no suffix) while `pulls/{n}/reviews` reports `copilot-pull-request-reviewer[bot]` |
Use the type fields, not the login suffix.
## Edge cases
- **Outdated threads** (`isOutdated: true`): line moved or was deleted;
`line` is often null → use `originalLine`. Default action: reply
"addressed in {commit}, line has since moved" + resolve.
- **Multi-round reviews**: re-running is idempotent — the unresolved-only
filter drops handled threads; GitHub's `isResolved` IS the state.
- **Summary-only feedback** (CHANGES_REQUESTED with no inline threads):
can't be resolved; reply as conversation comment and/or change code —
the review state flips on re-request.
- **No threads / all resolved**: report "No unresolved review threads on
PR #N", but still surface CHANGES_REQUESTED summaries.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |