Sentry Issue Analyzer

Perform read-only analysis of Sentry error monitoring data to inspect issues, retrieve events, and assess production application health through the Sentry API.

Overview

This skill queries the Sentry API to analyze production error data. It is designed for developers and SREs who need to investigate issues, understand error trends, and triage production incidents without leaving their development environment.

What it analyzes:

• •Open and unresolved issues across projects and environments
• •Individual issue details including frequency, first/last seen timestamps, and affected users
• •Event-level data including culprit, environment, release, and request context
• •Error trends over configurable time windows (1h to 90d)

What it reports:

• •Ranked issue lists with title, short ID, status, frequency, and recency
• •Issue detail breakdowns with metadata, tags, and assignment status
• •Event timelines with timestamps, environments, and release versions
• •Triage recommendations based on frequency, recency, and user impact

Security constraints:

• •All API calls are read-only (GET requests only)
• •PII (emails, IP addresses) is automatically redacted in output
• •Auth tokens are never echoed or logged
• •Stack trace entries are excluded by default (opt-in with --include-entries)

Authentication Setup

Before running any commands, the user must have a valid SENTRY_AUTH_TOKEN. If the token is not set:

1. •Direct the user to create one at: https://sentry.io/settings/account/api/auth-tokens/
2. •Required scopes: project:read, event:read, org:read
3. •Set the environment variable:
   bash

   export SENTRY_AUTH_TOKEN="sntrys_..."
   

4. •Optionally set defaults:
   bash

   export SENTRY_ORG="my-org"
   export SENTRY_PROJECT="my-project"
   export SENTRY_BASE_URL="https://sentry.io"  # or self-hosted URL
   

Never ask the user to paste the token directly in chat. Ask them to set it as an environment variable and confirm when ready.

Analysis Checklist

Issue Triage

• • List unresolved issues sorted by most recent occurrence
• • Identify high-frequency issues (count > 100 in 24h)
• • Check issue age (first_seen) to distinguish new regressions from chronic problems
• • Review assigned vs unassigned issues for ownership gaps
• • Compare issue counts across environments (prod vs staging)

Event Investigation

• • Retrieve latest events for a specific issue
• • Check event distribution across releases to identify regression sources
• • Review event context (browser, OS, URL) for pattern detection
• • Examine culprit field to locate the failing code path
• • Optionally inspect stack trace entries for root cause analysis

Health Assessment

• • Count total unresolved issues per project
• • Identify issues with rapidly increasing event counts
• • Check for issues affecting multiple environments
• • Review issues by tag to find systemic patterns (e.g., same browser, same endpoint)

Metrics

Issue Severity

Response Targets

Workflow

Step 1: Verify authentication

Confirm SENTRY_AUTH_TOKEN is set and org/project are configured:

SKILL_DIR="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")/.." 2>/dev/null && pwd || echo "$HOME/.claude/skills/sentry")"
python3 "$SKILL_DIR/scripts/sentry_api.py" \
  list-issues \
  --org "$SENTRY_ORG" \
  --project "$SENTRY_PROJECT" \
  --limit 1

If this fails, guide the user through authentication setup.

Step 2: List recent issues

python3 "$SKILL_DIR/scripts/sentry_api.py" \
  list-issues \
  --org "$SENTRY_ORG" \
  --project "$SENTRY_PROJECT" \
  --environment prod \
  --time-range 24h \
  --limit 25 \
  --query "is:unresolved"

Step 3: Investigate a specific issue

Resolve a short ID (e.g., PROJ-123) to a numeric issue ID:

python3 "$SKILL_DIR/scripts/sentry_api.py" \
  list-issues \
  --org "$SENTRY_ORG" \
  --project "$SENTRY_PROJECT" \
  --query "PROJ-123" \
  --limit 1

Then get full details:

python3 "$SKILL_DIR/scripts/sentry_api.py" \
  issue-detail \
  1234567890

Step 4: Examine events

List recent events for an issue:

python3 "$SKILL_DIR/scripts/sentry_api.py" \
  issue-events \
  1234567890 \
  --limit 10

Get a single event's detail:

python3 "$SKILL_DIR/scripts/sentry_api.py" \
  event-detail \
  --org "$SENTRY_ORG" \
  --project "$SENTRY_PROJECT" \
  abcdef1234567890

Add --include-entries only when the user explicitly requests stack trace data.

Step 5: Synthesize findings

After gathering data, present a structured analysis:

• •Summarize the top issues by severity and frequency
• •Identify patterns (same culprit, same release, same environment)
• •Recommend triage priorities based on the metrics table above
• •Suggest next steps (assign owner, link to code, escalate)

Report Format

Summary

Sentry Analysis Report
======================
Organization: {org}
Project: {project}
Time Window: {time_range}
Environment: {environment}
Generated: {timestamp}

Overview:
  Total unresolved issues: {count}
  New issues (< 24h): {count}
  High-frequency issues (> 100 events): {count}

Top Issues:
  1. {title} ({short_id}) - {count} events, last seen {last_seen}
  2. {title} ({short_id}) - {count} events, last seen {last_seen}
  3. {title} ({short_id}) - {count} events, last seen {last_seen}

Detailed Issue Report

Issue: {title}
Short ID: {short_id}
Status: {status}
First Seen: {first_seen}
Last Seen: {last_seen}
Events (24h): {count}
Users Affected: {user_count}
Assigned To: {assignee or "Unassigned"}
Culprit: {culprit}
Tags: {top_tags}

Recent Events:
  - {timestamp} | {environment} | {release} | {culprit}
  - {timestamp} | {environment} | {release} | {culprit}

Recommendation: {priority_level} - {suggested_action}

API Reference

All endpoints are read-only (GET). The bundled script handles authentication, pagination, and error retry.

Default Parameters

Examples

Example 1: Production incident triage

User: What are the top issues in production for the last 24 hours?
Assistant: Let me query Sentry for recent production issues.

[Runs list-issues with --time-range 24h --environment prod --limit 10]

Sentry Analysis Report
======================
Project: my-project | Environment: prod | Window: 24h

Top 5 Unresolved Issues:
1. TypeError: Cannot read property 'id' of undefined (PROJ-847)
   - 342 events | 89 users | Last seen: 12 min ago
   - Culprit: app/api/handlers/order.py

2. ConnectionError: Redis connection refused (PROJ-901)
   - 156 events | N/A | Last seen: 3 min ago
   - Culprit: app/cache/redis_client.py

3. ValidationError: Invalid email format (PROJ-512)
   - 87 events | 43 users | Last seen: 1h ago
   - Culprit: app/forms/registration.py

Recommendations:
- PROJ-847: Critical - investigate the undefined 'id' in order handler
- PROJ-901: High - check Redis connectivity, possible infrastructure issue
- PROJ-512: Medium - add input sanitization to registration form

Example 2: Specific issue investigation

User: Investigate PROJ-847 in more detail.
Assistant: Resolving PROJ-847 to get the full issue details...

[Runs list-issues with --query "PROJ-847" then issue-detail and issue-events]

Issue Detail: TypeError: Cannot read property 'id' of undefined
================================================================
Status: unresolved
First Seen: 2026-02-03T14:22:00Z (new regression)
Last Seen: 2026-02-04T15:43:00Z
Events (24h): 342
Users Affected: 89
Release: v2.4.1 (deployed yesterday)

Recent Events (last 5):
  - 15:43 | prod | v2.4.1 | app/api/handlers/order.py
  - 15:41 | prod | v2.4.1 | app/api/handlers/order.py
  - 15:38 | prod | v2.4.1 | app/api/handlers/order.py

Analysis:
- This is a new regression introduced in v2.4.1 (deployed yesterday)
- All events come from the same code path in order.py
- 89 unique users affected suggests it hits a common flow
- Recommendation: Roll back v2.4.1 or hotfix the order handler

Interpretation Guide

Issue Status

Time Range Selection

Output Rules

• •Always order issues by most recent occurrence
• •Show title, short_id, status, event count, last_seen, and culprit
• •Redact PII (emails, IPs) automatically via the bundled script
• •Never print raw auth tokens or full stack traces unless explicitly requested
• •If no results found, state explicitly: "No issues found matching the query"