DevOps & EngineeringIn Production

Sentry Auto-Fix

Polls Sentry for production errors, generates fixes with frontier models, and opens PRs automatically

In production: Monitors ClawsMarket's own production errors. Runs every 6 hours, has auto-fixed real bugs in this codebase.

An automated production bug fixer that integrates with Sentry to detect errors, uses Claude Opus to analyze stack traces and generate minimal code fixes, and creates GitHub pull requests with the patches.

The pipeline runs on a schedule (every 6 hours by default), polling your Sentry project for unresolved issues. For each new error, it fetches the full stack trace and breadcrumbs, reads the relevant source files from your local checkout, and sends everything to Claude for analysis.

Claude produces targeted, minimal fixes — no refactoring, no style changes, just the bug fix. Each fix includes a confidence rating (high/medium/low) and an explanation. The pipeline then creates a branch, applies the fix, pushes, and opens a PR with the Sentry link and diff.

Includes a local dashboard for tracking issues, fixes, PRs, and pipeline runs.

Your agent controls the pipeline via API — adjust confidence thresholds, ignore specific error levels, swap models, and set schedules.

What's included

Automated Sentry issue polling and deduplication

Full stack trace and breadcrumb analysis

Source file context gathering from local checkout

Minimal, targeted code fixes via Claude Opus

Confidence-gated PR creation

Automatic branch, commit, and GitHub PR workflow

Local tracking dashboard

Email notifications on pipeline runs

Full config API — your agent controls everything

Agent Config API

Everything is editable via API — prompts, templates, scoring rubrics, model choices, schedules. Your agent reads the config, changes what it needs, and triggers runs. No source code needed.

Prompts & Logic— Edit the AI's instructions, scoring criteria, and decision-making rules

fix_promptmultiline

System prompt for the bug fixer model. Controls how it analyzes errors, formats fixes, and decides when to skip.

triage_promptmultiline

Instructions for triaging errors before analysis. Controls which errors to prioritize or ignore.

Thresholds & Tuning— Adjust sensitivity, limits, and quality gates

max_issues_per_runnumber

Maximum number of new issues to analyze per pipeline run. Default: 10.

min_confidencechoicelow | medium | high

Minimum confidence level to create a PR. Lower confidence fixes are stored but not pushed.

lookback_hoursnumber

How far back to look for unresolved issues. Default: 168 (7 days).

ignore_levelsJSON

Error levels to skip. Default: ["debug", "info"].

Models— Choose which AI models power each stage

fix_modelchoiceclaude-opus | claude-sonnet | gpt-4o

Frontier model for generating code fixes.

Scheduling— Control when the pipeline runs

schedule_croncron

When the pipeline runs. Default: every 6 hours.

Integrations— Connect your tools and services

sentry_projecttext

Sentry project slug to monitor.

sentry_orgtext

Sentry organization slug.

github_repotext

GitHub repo in owner/name format (e.g. myorg/myapp).

github_base_branchtext

Branch to create fix PRs against. Default: main.

webhook_urltext

URL for real-time notifications (new issues, PRs created).

bash

# Read full config (all prompts, templates, thresholds, etc.)
curl https://www.clawsmarket.com/api/workflows/sentry-auto-fix/config \
  -H "Authorization: Bearer claws_YOUR_KEY"

# Edit a prompt — your agent rewrites the AI's instructions
curl -X PATCH https://www.clawsmarket.com/api/workflows/sentry-auto-fix/config \
  -H "Authorization: Bearer claws_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "personalization_prompt": "Write casual, founder-to-founder icebreakers...",
    "scoring_threshold": 15,
    "schedule_cron": "0 2 * * *"
  }'

# Replace email templates with your own
curl -X PATCH https://www.clawsmarket.com/api/workflows/sentry-auto-fix/config \
  -H "Authorization: Bearer claws_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "email_templates": { "step_1": [...], "step_2": [...] } }'

# Reset everything to battle-tested defaults
curl -X POST https://www.clawsmarket.com/api/workflows/sentry-auto-fix/reset \
  -H "Authorization: Bearer claws_YOUR_KEY"

# Trigger a run now
curl -X POST https://www.clawsmarket.com/api/workflows/sentry-auto-fix/trigger \
  -H "Authorization: Bearer claws_YOUR_KEY"

# Get latest outputs
curl https://www.clawsmarket.com/api/workflows/sentry-auto-fix/outputs \
  -H "Authorization: Bearer claws_YOUR_KEY"

Tools used

AgentMail

Skills used

Email Notifications

Tech stack

SQLitePythonFlaskGitHub CLI

Community

0 upvotes

7-day free trial

$0/mo

Full API access — no feature gates
Dashboard + agent config API
All integrations included
Cancel anytime

Support included

Phone support

Live chat