getsentry / sentry

About half are guides to Sentry's own internal architecture: cell-architecture explains its multi-region cell and locality model and the live migration off region naming; hybrid-cloud-rpc, hybrid-cloud-outboxes, and hybrid-cloud-test-gen cover the RPC and eventual-consistency machinery that architecture requires; django-models and generate-migration are about placing new data correctly inside that same silo system. A second group enforces frontend convention rather than building anything new — design-system, cmdk-actions, generate-frontend-forms, migrate-frontend-forms, lint-new, and lint-fix all exist to make a change match a house style. A third is code review tuned to this codebase's own incident history: sentry-backend-bugs, sentry-javascript-bugs, and sentry-security each describe patterns their own descriptions say are drawn from real production issues, not generic advice.

This is for someone opening a pull request against getsentry/sentry, not for a team that has installed Sentry to catch its own errors — that team will find nothing here about SDKs, alerting, or self-hosting. Fifteen of the twenty need nothing beyond the skill file itself, which fits their nature as internal conventions rather than integrations; the rest, setup-dev especially, assume you can already run devenv and the rest of Sentry's local development stack.

READ THE FULL ANALYSIS＋

The readme doesn't mention any of this. getsentry/sentry's actual readme is what a public product's readme looks like: what Sentry is, links to its SDKs across languages, docs, and Discord. It says nothing about skills or an agent. We read the repository directly; nothing here is a claim Sentry itself makes about the package.

One of the 20 talks to an external service. Our data marks one skill mcp_service rather than a plain local guide, meaning it expects an MCP connection rather than only reading the codebase in front of it — worth checking before assuming every skill here is offline-only.