Labsco
Labsco/Repos/getsentry/sentry-for-ai
REPO PACKAGE

getsentry sentry-for-ai

OFFICIAL
getsentry · publisher268 repository starsMITgithub.com/getsentry/sentry-for-ai
32skills
3groups
3ready to use

The skills do three different jobs. Most are full SDK setup wizards — one per platform, each scanning the project and walking through error monitoring, tracing, profiling, session replay, and logging for that stack; the README's own table names fifteen platforms, but the live skill list also includes wizards for the browser, Cloudflare Workers, Elixir, and Flutter that the table leaves out. A smaller group covers specific features: AI/LLM call monitoring, an OpenTelemetry Collector exporter, and a guide for choosing between an error, a span, a log, or a metric. The rest is workflow — finding and fixing issues through the Sentry MCP server, creating alerts, and generating new SDK-wizard skills from a template.

This is for a team already sending errors to Sentry, or about to, that wants its coding agent to install the SDK correctly the first time and later triage what Sentry catches without leaving the editor — not for evaluating whether to adopt Sentry itself. The distribution model matters too: this repository is the source of truth but is not itself installable; four separate downstream repositories, one each for Claude Code, Cursor, Codex, and Grok, are built from it by CI and are what actually gets added to an agent.

READ THE FULL ANALYSIS

What it costs to start. Twenty-four of the 33 skills need a live Sentry project — and the plugin's auto-configured MCP connection — before they do more than write install code. A few workflow skills also expect the GitHub CLI (gh) authenticated, which the README calls out but does not set up for you.

Who this isn't for. If you don't run Sentry, none of this applies: every skill here assumes an existing or soon-to-exist Sentry project, and there's no offline or eval mode the way a purely local tool would have one.

The README undercounts its own wizards. Its "SDK Setup Wizards" table lists fifteen platforms, which is the number worth quoting if you're citing this package, but four more wizard skills exist in the live set than that table shows — a small but real gap between the docs and what actually ships.

32Grouped into 3 sets by the authors.
3Work with nothing else to set up.
268Stars on the GitHub repository, at last check.