sentry

name: sentry description: Fetch active Sentry issues for the Personas project, diagnose root causes from stack traces, apply code fixes, and mark issues as resolved. Covers both Rust backend (tracing::error!, panics) and React frontend (exceptions, unhandled rejections). allowed-tools: Read, Write, Edit, Grep, Glob, Bash

Sentry Issue Triage & Fix Pipeline

You process unresolved Sentry issues for the personas project in org d3v-center. For each issue: fetch the stack trace, locate the offending source, propose a fix, apply it after user approval, verify it compiles, commit, and mark the issue resolved in Sentry via the API.

Constants

• Sentry org: d3v-center
• Sentry project: personas
• Auth: SENTRY_AUTH from .env at repo root (user auth token, sntryu_* prefix)
• API base: https://sentry.io/api/0
• Triage output: .claude/sentry/TRIAGE-<YYYYMMDD>.md

Trigger

• /sentry — list all unresolved issues, pick one or a batch to fix
• /sentry list — list only, no fixes
• /sentry fix <issue-id> — go straight to fixing a specific issue
• /sentry top <N> — triage the N issues with the highest event count (default 5)
• /sentry stats — summary: unresolved count, frequency buckets, affected releases

Coordination — Active-Runs Ledger

Before applying any code fix or marking a Sentry issue resolved, register this session in .claude/active-runs.md per the convention in CLAUDE.md → Concurrent CLI sessions. Read the file's ## Active section first; if any started-status entry overlaps your planned scope and is <2h old, surface the conflict to the user before proceeding. Overlap on .claude/active-runs.md itself is expected and is not a conflict.

Declared paths for /sentry: scope is per-issue, not fixed up front. After Phase 1 fetches issues and the user picks one or a batch, declare:

• The Rust/TS files referenced in each chosen issue's stack trace (typically src-tauri/src/**/*.rs and src/**/*.{ts,tsx} files at the top of the trace)
• .claude/sentry/TRIAGE-<YYYYMMDD>.md (the triage report)
• Always: .claude/active-runs.md

If batch-fixing N issues that touch overlapping files, register ONE entry covering the union of paths (not N entries) — the ledger reads as "this session is in flight on those files".

At session end (after the final git commit for the last fix in the batch + Sentry resolve API call): move your entry to the top of ## Recently completed. Update Status to completed (commit: <last-sha>) (or list multiple SHAs if batch) or aborted (<reason>). Trim entries older than 14 days while you're there.

Full design rationale: docs/concepts/cli-coordination-active-runs.md.

Parallel-safety primitives (mandatory)

Per CLAUDE.md → Parallel-safety primitives, every CLI session must:

1. Never git stash other sessions' work — not even with --keep-index. If your commit step needs a clean stage, use git add <path> per file (NOT git add -A / git add . / git add -u); leave everything else alone.
2. Use a worktree for multi-issue batches. Single-issue fixes that touch ≤1 file can stay on the main checkout. For batch mode (/sentry top N, multiple issues, or any fix touching multiple files):

   git worktree add .claude/worktrees/sentry-<YYYYMMDD> -b worktree-sentry-<YYYYMMDD>
   cd .claude/worktrees/sentry-<YYYYMMDD>
   

3. Atomic commits per issue. One commit per Sentry issue resolution — never bundle N issue fixes into one commit. The Sentry issue ID belongs in the commit message subject (fix(sentry): <issue-id> — <short title>) so the resolve API call and the commit are traceable from each other.
4. Verify the staged index before commit. After git add and before git commit, run git diff --cached --stat. If the staged file count is greater than the number you explicitly added, another session pre-staged work in the index — git restore --staged <path> per unrelated file, or use git commit --only <files> to bypass the shared index entirely.
5. Clean up the worktree after merge. Once the worktree's branch is in git log master, from the main checkout: git worktree remove .claude/worktrees/sentry-<YYYYMMDD> and git branch -D worktree-sentry-<YYYYMMDD>. Treat as part of the session-end ledger ritual.

Phase 0: Load auth and smoke-test the API

TOKEN=$(grep -E "^SENTRY_AUTH=" .env | cut -d= -f2-)
if [ -z "$TOKEN" ]; then
  echo "ERROR: SENTRY_AUTH not set in .env"
  exit 1
fi

# Smoke test — confirm token is valid and project is reachable
curl -sS -H "Authorization: Bearer $TOKEN" \
  "https://sentry.io/api/0/projects/d3v-center/personas/" \
  | python -c "import sys,json; d=json.load(sys.stdin); print(f\"Project OK: {d.get('slug')} — {d.get('eventProcessing',{}).get('symbolicationDegraded','symbolic ok')}\")"

Abort if the smoke test fails (bad token, project renamed, org changed).

Ensure .claude/sentry/ exists: mkdir -p .claude/sentry.

Phase 1: Fetch unresolved issues

curl -sS -H "Authorization: Bearer $TOKEN" \
  "https://sentry.io/api/0/projects/d3v-center/personas/issues/?query=is:unresolved&statsPeriod=14d&limit=100&sort=freq" \
  > .claude/sentry/issues.json

Parse with python/jq. Fields to keep per issue:

Filter out noise before presenting:

• Skip level: warning unless user passed --include-warn
• Skip issues not seen in the last 7 days (stale — probably already fixed)
• Skip issues with count == 1 AND lastSeen > 3 days ago (likely flukes)

Phase 2: Present triage table — one approval gate

Write .claude/sentry/TRIAGE-<date>.md:

# Sentry Triage — 2026-04-17

Total unresolved: 42
Filtered (active, >7d): 18

| # | shortId | lvl | count | users | lastSeen | platform | title |
|---|---------|-----|-------|-------|----------|----------|-------|
| 1 | PER-1A  | err | 847   | 23    | 2h ago   | rust     | ONNX Runtime panicked: LoadLibraryExW failed |
| 2 | PER-2B  | err | 412   | 17    | 4h ago   | js       | TypeError: Cannot read properties of undefined (reading 'status') |
| ...

Show the user this table and ask:

Which do you want to fix? Reply with indexes (1,2,5), all, top 3, or skip.

Do not proceed without explicit user selection. This is the single approval gate.

Phase 3: Fix each selected issue

For each issue (iterate sequentially — fixes may depend on each other):

3.1 Pull the latest event with full stack

ISSUE_ID="<numeric id>"
curl -sS -H "Authorization: Bearer $TOKEN" \
  "https://sentry.io/api/0/issues/$ISSUE_ID/events/latest/" \
  > .claude/sentry/event-$ISSUE_ID.json

Extract:

• Exception class + message: entries[].data.values[0].type + .value
• Frames: entries[].data.values[0].stacktrace.frames[] — each has filename, function, lineNo, colNo, inApp, context
• Breadcrumbs: entries[] | select(.type=="breadcrumbs") | .data.values[] — UI/state just before the crash
• Tags: tags[] — release, OS, environment
• Request (frontend only): entries[] | select(.type=="request") | .data.url

Focus on inApp: true frames — those are our code. Skip library internals.

3.2 Locate the source

Map Sentry paths to local paths. Common patterns:

Read the file at the reported line. If source maps are missing for a frontend issue, grep for a distinctive string in the exception message.

3.3 Diagnose

Write a short diagnosis (4-6 bullet points) covering:

• What the user was doing (from breadcrumbs + request.url for frontend)
• Why the crash happens (null deref, race, missing guard, type mismatch, etc.)
• Whether it's a real bug or expected failure that needs better handling (e.g., network down)
• Scope: is this a one-line guard or a structural fix?

If the issue is not actionable (hardware failure on one user's machine, third-party API flake, dev-only code path), mark for resolution with justification and skip code changes — jump to Phase 3.6 with a resolution: "wontfix" body.

3.4 Apply the fix

Use Edit directly on the source. Style guidance (inherits from CLAUDE.md):

• No extraneous comments; only add one if the WHY is non-obvious
• For user-facing frontend text, respect i18n (go through t.*, never hardcode English)
• For backend errors, prefer tracing::error! with structured fields over bare anyhow::bail!
• No backwards-compat shims — fix it cleanly

Verify locally:

# If frontend-only:
npx tsc --noEmit                              # must be 0 errors
# If backend:
cd src-tauri && cargo check 2>&1 | tail -30

Do not resolve the Sentry issue until the check passes.

3.5 Commit

One commit per issue, referencing the shortId:

git add <fixed files>
git commit -m "$(cat <<'EOF'
fix(<area>): <one-line summary>

Sentry: <shortId> — <permalink>
Root cause: <1-2 sentences>
Fix: <1 sentence describing the change>

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EOF
)"

Follow the repo's commit style (lowercase conventional commits, see git log --oneline -10).

3.6 Mark resolved in Sentry

curl -sS -X PUT -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  "https://sentry.io/api/0/issues/$ISSUE_ID/" \
  -d '{"status":"resolved","statusDetails":{"inNextRelease":true}}'

inNextRelease: true means Sentry auto-reopens the issue if it recurs after the next release is deployed — that's what we want for a code fix that ships in a future CI build. For wontfix, use {"status":"ignored"} instead.

Record in the triage file:

## PER-1A — Fixed ✅
- Commit: abc1234
- Files: src/features/.../Foo.tsx
- Strategy: Added null guard around `session.metadata`

Phase 4: Summary report

After all selected issues are processed, append to the triage file:

## Summary — 2026-04-17 15:42

- Issues triaged: 18
- Fixed and resolved: 12
- Ignored (wontfix): 2
- Skipped (needs design decision): 4
- Commits created: 12
- Files changed: 19

Print this summary to the console. Done.

Safety rails

1. Never resolve an issue you didn't fix. If the user says "skip this one", leave the Sentry status untouched.
2. Never commit without the sanity check passing (tsc --noEmit for frontend, cargo check for backend).
3. Never bulk-resolve without individual inspection — Sentry's API can resolve in bulk, but that defeats the purpose.
4. Never fetch more than 100 issues per call — paginate via the Link response header if needed.
5. Don't log the PAT. Read it once into $TOKEN, never echo it. If the user shares the transcript, they shouldn't see it.
6. Respect CLAUDE.md "Do Not Fix Unless Asked" list — if a Sentry issue points at a file on that list, mark it for user triage rather than auto-fixing.

References

• Sentry API docs: https://docs.sentry.io/api/
• Issue status values: resolved, unresolved, ignored
• Project error reporting config: docs/devops/guide-error-reporting.md
• PII scrubbing: src-tauri/src/main.rs (Rust before_send), src/lib/sentry.ts (JS beforeSend)