sentry-python-sdk

An opinionated setup wizard for adding the Sentry SDK to a Python application, covering error monitoring, tracing, profiling, logging, metrics, crons, and AI monitoring. It applies to requests such as adding Sentry to Python, installing sentry-sdk, or configuring error monitoring in a Python app, and it supports many frameworks including Django, Flask, FastAPI, Celery, Starlette, AIOHTTP, and Tornado.

Work proceeds in three phases. Detect runs a series of grep commands to understand the project before recommending anything, checking whether sentry-sdk is already present, which web framework and task queue are used, which logging and AI libraries are installed, whether schedulers or cron patterns exist, and whether OpenTelemetry tracing is in play. Recommend leads with a concrete proposal rather than open-ended questions, driven by a feature matrix: error monitoring is always recommended as the baseline, tracing when an HTTP framework is detected, AI monitoring when libraries such as OpenAI, Anthropic, or LangChain are present and auto-instrumented with zero config, and profiling, crons, and metrics when their triggers appear. Guide then handles installation, a recommended init configuration, and where to place that init call for each framework.

A notable branch is OpenTelemetry handling: when OTel tracing is detected, the skill routes to an OTLP integration path, deliberately does not set a traces sample rate, and skips profiling because it is incompatible with that path, letting Sentry link errors to existing OTel traces automatically. Per-framework guidance specifies exactly where initialization belongs, for example at the top of Django settings, before app creation in Flask and FastAPI, and dual-process init for Celery. The skill notes that its API details reflect sentry-sdk 2.x at time of writing and advises verifying against the current Python platform docs before implementing. Detailed configuration for each feature lives in referenced files. It is part of a broader Sentry SDK setup skill tree and is marked so it is not invoked by the model automatically.