fix-release-pr

作者: clickhouse

Fix CI failures and address code-review comments on a pull request that targets the protected `release` branch of `ClickHouse/clickhouse-js`. Release PRs are…

npx skills add https://github.com/clickhouse/clickhouse-js --skill fix-release-pr

Fixing PRs to the release branch

Why this skill exists

In clickhouse-js, the release branch receives release PRs — snapshots of main opened to cut a version (e.g. titled "1.23 beta2"). Two properties make them special:

  • The PR's head branch is usually main itself (base release, head main). gh pr checkout <N> is therefore a no-op that leaves you on main — do not commit fixes there.
  • release is protected: you cannot push commits onto the release PR to fix CI or review feedback.

So fixes never go onto the release PR. They go to main via a separate PR; once that merges, the release branch is re-synced from main and the release PR picks the fix up. This skill is that workflow.

Step 1 — Confirm it's a release PR

gh pr view <N> --json title,baseRefName,headRefName,headRefOid

If baseRefName is release, proceed. (If it's main, this skill does not apply — push to the PR's branch normally.)

Step 2 — Gather what needs fixing

CI failures. List checks and find the fail rows:

gh pr checks <N>
  • The job named success is an aggregate gate ("Fail if any needed job failed") — it only fails because a real job failed. Ignore it as a root cause and find the actual failing job.
  • Open the real failure log:
gh run view --job <JOB_ID> --log-failed | tail -50

Review comments. Inline review comments with their REST IDs and the GraphQL thread node IDs (you need both: REST id to reply, thread node id to resolve):

# Inline comments: REST id + location + author + body
gh api repos/ClickHouse/clickhouse-js/pulls/<N>/comments \
  -q '.[] | "\(.id)\t\(.path):\(.line)\t\(.user.login)\n\(.body)\n---"'

# Review threads: node id (PRRT_…), resolved state, and first comment's databaseId
gh api graphql -f query='
{ repository(owner:"ClickHouse", name:"clickhouse-js") {
    pullRequest(number: <N>) {
      reviewThreads(first: 50) { nodes {
        id isResolved
        comments(first: 1) { nodes { databaseId path body } }
      } }
    } } }' \
  -q '.data.repository.pullRequest.reviewThreads.nodes[]
      | "\(.id)\tresolved=\(.isResolved)\tdbId=\(.comments.nodes[0].databaseId)\t\(.comments.nodes[0].path)"'

Match each thread (PRRT_… node id) to its first comment's databaseId — that databaseId is the REST comment id you reply to in Step 5.

Step 3 — Branch off the latest main

Never branch off the release PR head. Start from up-to-date main:

git fetch origin
git checkout -b fix/<short-topic> origin/main

Step 4 — Make the fix and verify

Apply the fixes. Then verify with the repo's own tooling (run the setup skill first if node_modules isn't populated):

  • Prettier is the most common release-PR CI failure. House style is the Prettier defaults (.prettierrc is {} → double quotes + semicolons). Fix and check:
    node_modules/.bin/prettier --write <file>
    npm run -s prettier:check
    
    prettier:check runs on the whole repo and may flag untracked local scratch dirs (e.g. a type-parser/ working directory). Those aren't part of the PR and CI never sees them — only tracked files matter. Confirm no tracked file is flagged:
    npm run -s prettier:check 2>&1 \
      | grep -oE '\[(warn|error)\] [^ ]+\.(mjs|ts|js|json|ya?ml|md)' \
      | grep -v '<your-untracked-dir>/' | sort -u   # empty = clean
    
  • For standalone scripts: node --check <file>.
  • For library code: npm run typecheck, npm run lint, and the relevant npm run test:* target (see the setup skill for what each needs).
  • When the fix is logic (not just formatting), exercise it directly — e.g. drive the script against synthetic inputs in the scratchpad dir and assert the exit code / output, rather than trusting it by inspection.

Step 5 — Commit, push, open the PR to main

git commit -m "<type>(<scope>): <summary>

Addresses CI failure / review feedback on #<N> (release PR). The fix lands
on \`main\` separately because #<N> targets the protected \`release\` branch.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"

git push -u origin fix/<short-topic>
gh pr create --base main --head fix/<short-topic> --title "…" --body "…"

In the new PR body, explicitly state it addresses #N and why it's a separate PR (release branch is protected). Map each fix back to the specific CI failure or review comment it resolves.

Step 6 — Close the loop on the release PR

For each original review comment, reply pointing to the new PR, then resolve the thread.

# Reply (use the REST comment id from Step 2)
gh api repos/ClickHouse/clickhouse-js/pulls/<N>/comments/<COMMENT_ID>/replies \
  -f body='Fixed in #<NEW_PR>. <one line on what changed>. (Lands on `main` separately since this PR targets the protected `release` branch.)' \
  -q '.html_url'

# Resolve the thread (use the PRRT_… node id from Step 2)
gh api graphql \
  -f query='mutation($id:ID!){resolveReviewThread(input:{threadId:$id}){thread{id isResolved}}}' \
  -f id='<THREAD_NODE_ID>' \
  -q '.data.resolveReviewThread.thread | "\(.id) resolved=\(.isResolved)"'

Step 7 — Flag the re-sync

The fix is on main, not on release. Remind the maintainer that once the new PR merges, the release branch / the release PR must be re-synced from main to pick the fix up. Do not attempt to push to release yourself.

Quick reference — the whole flow

  1. gh pr view <N> --json baseRefName → confirm base is release.
  2. gh pr checks <N> + gh run view --job <id> --log-failed → real CI failure (ignore the success gate).
  3. gh api …/pulls/<N>/comments + GraphQL reviewThreads → review comments + thread ids.
  4. git checkout -b fix/… origin/main → branch off latest main.
  5. Fix → verify (prettier / typecheck / lint / tests / node --check).
  6. Commit → push → gh pr create --base main.
  7. Reply to each review comment (REST …/comments/<id>/replies) → resolve each thread (GraphQL resolveReviewThread).
  8. Remind: re-sync release from main after merge.

来自 clickhouse 的更多技能

clickhouse-best-practices
clickhouse
28条ClickHouse最佳实践规则,按模式设计、查询优化和数据摄入策略组织。涵盖三个关键领域:主键与数据类型选择(不可变设计决策)、JOIN与查询优化、批量插入与避免突变。包含28条按影响程度排序的规则,其中模式设计和查询优化规则因ClickHouse的列式存储和稀疏索引机制被标记为关键。提供结构化审查流程用于...
official
clickhouse-js-node-coding
clickhouse
Write idiomatic application code with the ClickHouse Node.js client (`@clickhouse/client`). Use this skill whenever a user is *building* against the Node.js…
official
clickhousectl-cloud-deploy
clickhouse
当用户希望将ClickHouse部署到云端、投入生产环境、使用ClickHouse Cloud、托管托管式ClickHouse服务,或从本地迁移时使用…
official
clickstack-otel-collector
clickhouse
当用户希望将OpenTelemetry collector接入ClickHouse Cloud上的托管ClickStack服务时使用,无论是通过部署新的本地collector…
official
infra-clickhouse
clickhouse
使用 clickhousectl CLI 设置并管理 ClickHouse — 安装并运行本地 ClickHouse 服务器用于开发,以及创建托管的 ClickHouse Cloud…
official
infra-postgres
clickhouse
Sets up and manages Postgres using the clickhousectl CLI — runs a local Docker-backed Postgres for development, and creates and operates managed ClickHouse…
official
clickhouse-best-practices
clickhouse
在审查ClickHouse模式、查询或配置时必须使用。包含31条在提供建议前必须检查的规则。始终阅读…
official
setup
clickhouse
引导用户完成此插件附带的ClickHouse MCP服务器连接的设置。在用户首次安装插件或遇到问题时使用…
official