sentry-triage

name	sentry-triage
description	Pull all unresolved issues from Sentry (bakabase.sentry.io), triage each, reuse or create concise Chinese GitHub issues with labels, fix them on a feature branch, and open a PR that auto-closes those issues. Use when the user asks to process, triage, or fix Sentry issues / production errors.

Sentry Issue Triage Workflow

End-to-end batch workflow: Sentry unresolved issues → analysis → Chinese GitHub issues → fixes → PR.

This skill is the authoritative process for batch-handling Sentry issues. It intentionally supersedes the generic per-issue "自动化 Issue 处理规范" 4-step flow in .claude/CLAUDE.md: here you proceed automatically for issues that are fixable and small/medium scope, and only pause to ask the user about issues that are unfixable or very large in scope.

Tooling — do not assume any specific CLI

This workflow may run locally or in a cloud agent. Each step states an intent; achieve it with whatever the environment provides:

• HTTP (Sentry API) — curl, or any built-in fetch capability.
• GitHub (issues, PRs) — the gh CLI if installed; otherwise the GitHub REST API or the authorized GitHub integration available to the agent.
• git (branch, commit, push) — always available; it is a git repository.

Repository: anobaka/Bakabase. Default branch: main.

Language

• GitHub issues — title and body in Chinese, concise (简洁).
• Commit messages and PR title/description — in English.

Prerequisites

• BAKABASE_SENTRY_PAT — Sentry auth token, read from the environment. If it is not set, stop and tell the user.
• Run from the Bakabase/ git repository root.

Step 1 — Fetch unresolved Sentry issues

Sentry org slug: bakabase. API base: https://bakabase.sentry.io/api/0. Authenticate every request with the header Authorization: Bearer <BAKABASE_SENTRY_PAT>.

1. List projects: GET /organizations/bakabase/projects/.
2. For each project, list unresolved issues — follow Link-header cursor pagination until exhausted: GET /projects/bakabase/<project-slug>/issues/?query=is%3Aunresolved&sort=freq&limit=100
3. For each issue, fetch the latest event for its stack trace: GET /issues/<issueId>/events/latest/

From each issue keep shortId, title, culprit, level, count, userCount, permalink, metadata. From the latest event keep the exception entry (stack frames) and breadcrumbs.

Example with curl:

curl -sS -H "Authorization: Bearer $BAKABASE_SENTRY_PAT" \
  "https://bakabase.sentry.io/api/0/projects/bakabase/<project-slug>/issues/?query=is%3Aunresolved&sort=freq&limit=100"

Step 2 — Triage each issue

For every issue:

1. Read the title, culprit, level, and frequency (count / userCount).
2. Walk the exception stack frames; map in-app frames to local source files and read the relevant code.
3. Determine the root cause and a concrete fix plan.
4. Group issues that share one root cause — fix them together, with one GitHub issue per distinct root cause (or one per Sentry issue when unrelated).

Prioritise by impact (count × userCount, then level).

Step 3 — Classify: fix, or ask the user

• Fixable, small/medium scope → proceed (Steps 4–6); do not ask.
• Cannot be fixed (needs reproduction you cannot do, third-party/SDK bug, environment-only, insufficient data) or estimated to be a very large change → ask the user via AskUserQuestion whether to skip it. State the root cause briefly and why it is hard. Batch these questions when several issues qualify.

Issues the user chooses to skip: leave them untouched and list them in the final summary and the PR description.

Step 4 — Reuse or create a GitHub issue

Before creating anything, look for an existing open issue that already covers the same root cause — a previous run of this workflow, or the maintainer, may have already filed it. List or search open issues on anobaka/Bakabase (match on the Sentry shortId, the exception type, or the culprit) and read any candidate.

• A suitable open issue already exists → reuse it: record its number #N and skip creation. Do not open a duplicate; leave its body as-is.
• No suitable open issue → create one as described below.

An issue counts as "suitable" only when it describes the same root cause (or the same Sentry issue) and is still open — a closed issue does not count, file a fresh one. When uncertain, prefer reusing over creating.

Creating a new issue

Follow .claude/CLAUDE.md → "Workflow: GitHub Issue Management". Every issue MUST be 中文, 简洁, and labelled — pick existing repo labels (bug is the usual fit for Sentry crashes); create one only if none fits.

Issue body template (Chinese):

## 问题
<一句话描述错误现象>

## 根因
<简要根因分析>

## 修复
<简要修复方案>

## Sentry
<shortId> · <permalink>

Create it on anobaka/Bakabase — gh issue create --repo anobaka/Bakabase --title ... --body ... --label bug, or the equivalent GitHub API call. Record the returned issue number #N.

Step 5 — Fix on a feature branch

• Create one feature branch for the batch off main: git checkout -b fix/sentry-triage-<YYYYMMDD>.

• Implement the fixes. Split into multiple commits by logical unit — the number of commits is unconstrained.

• Build / verify per project conventions where feasible (dotnet build, frontend checks).

• Commit messages are in English; each body MUST contain a closing-keyword line per issue it resolves:

  fix: <concise English description>
  
  Fixes #N
  

  Fixes / Closes / Resolves are interchangeable. Do not add [skip ci] to code-fix commits.

Step 6 — Open the PR

Push the branch and open one PR for the batch (gh pr create or the GitHub API). Title and description in English. The description MUST also list every closing keyword — auto-close fires when the commits reach main via the merged PR:

## Summary
Batch fixes for unresolved Sentry issues.

## Closes
Closes #N1
Closes #N2

## Skipped
- <shortId>: <reason> (confirmed with the user)

Output the PR URL to the user.

Step 7 — (Optional) Resolve in Sentry

Only if the user asks, mark the corresponding Sentry issues resolved after the PR is merged:

curl -sS -X PUT \
  -H "Authorization: Bearer $BAKABASE_SENTRY_PAT" \
  -H "Content-Type: application/json" \
  -d '{"status":"resolved"}' \
  "https://bakabase.sentry.io/api/0/organizations/bakabase/issues/<issueId>/"

Final summary

Report back: issues fixed (with #N), issues skipped (with reason), and the PR URL.