sentry

Overview of sentry skill

The sentry skill is a read-only Observability tool for investigating Sentry data fast: issues, events, health signals, and likely production impact. It is best for people who want the answer to “what is breaking, where, and how bad is it?” without manually clicking through Sentry. Compared with a generic prompt, the sentry skill is tuned for CLI-based inspection, so it is more reliable when you need current data, repeatable queries, and structured output.

Use this skill when you already know you need Sentry-specific insight and want a practical sentry guide for triage, not a broad observability overview. It is a good fit for incident responders, engineers checking regressions, and analysts who need short, verifiable summaries from live Sentry data.

What the sentry skill does well

It supports read-only investigation through the sentry command, with defaults that favor quick triage: recent time windows, production-first filtering, and JSON output for downstream processing. The skill is especially useful when you need to list unresolved issues, inspect a specific issue by short ID, or pull structured data without losing pagination or auth handling.

When this skill is a good fit

Choose the sentry skill if your task depends on live Sentry state, you need reproducible queries, or you expect to iterate on filters like environment, project, time range, or priority. It is also a good fit when the user wants a sentry for Observability workflow that can be reused in automation or agent-based analysis.

What to know before installing

This skill is read-only. It does not create issues, change alerting, or modify project settings. Adoption is blocked most often by missing authentication, unclear org/project targeting, or trying to use it for non-Sentry data. If you need write actions or dashboards, this is not the right skill.

How to Use sentry skill

Install and authenticate

Install the sentry skill with npx skills add openai/skills --skill sentry. If the CLI is not already available, install Sentry CLI first, then authenticate locally with sentry auth login or SENTRY_AUTH_TOKEN. Confirm access with sentry auth status before asking for investigation work. Do not paste tokens into chat.

Give the skill the right input

For best sentry usage, ask for a concrete investigation goal plus a scope. Strong inputs name the symptom, environment, and time window, such as: “Investigate the latest production 5xx spike in the last 24h and summarize the top unresolved issues.” If you know the project, include it; if not, let auto-detection try first. The CLI can usually infer org/project from DSNs, source code, config defaults, and directory names.

Start with the right files and commands

Read SKILL.md first, then check agents/openai.yaml for the default behavior and prompt framing. Use sentry schema <resource> when you need to discover available endpoints or fields. For machine-readable results, always request --json, and combine it with --fields to limit noise and make summaries more accurate.

Practical workflow for better results

A solid sentry guide workflow is: confirm auth, verify target org/project, run a narrow list query, then inspect one or two issue details before summarizing. For example, start with unresolved production issues from the last 24h, then refine by release, environment, or priority if the first pass is too broad. This keeps the analysis grounded in current data instead of overfitting to one noisy event.

sentry skill FAQ

Does the sentry skill need Sentry CLI?

Yes. The sentry skill is designed around the sentry CLI, not a browser workflow. If the CLI is missing, install it and authenticate locally first. That setup is what makes the sentry install worthwhile for repeatable, read-only investigation.

How is this different from a normal prompt?

A normal prompt can describe a Sentry query, but the sentry skill adds a concrete operating path: auth expectations, default query conventions, JSON output, and field selection. That reduces guesswork and makes the result easier to verify or automate.

Is the sentry skill beginner friendly?

Yes, if the user can state a problem clearly. Beginners usually do well when they provide “what broke, when, and where” rather than trying to craft an exact CLI query. The skill handles the querying mechanics; the user only needs to define the investigation target.

When should I not use it?

Do not use the sentry skill for editing Sentry configuration, changing alert policies, or any workflow that needs write access. It is also a poor fit if you do not have access to the relevant org/project or if your question is about application logs outside Sentry.

How to Improve sentry skill

Provide sharper investigation context

The sentry skill works best when you specify environment, time range, and the symptom you care about. “Summarize recent errors” is weaker than “Find the top unresolved production errors in the last 6h after the 2.4.1 release.” Better context improves issue ranking, reduces false positives, and helps the agent choose the right filters.

Ask for structured output

If you want a useful sentry for Observability result, ask for a table or bullet summary with short ID, title, severity, and likely impact. That nudges the workflow toward actionable output instead of a vague narrative. If you need machine use, request --json-friendly fields so the response stays compact and parseable.

Iterate by narrowing, not broadening

If the first pass is noisy, narrow by release, environment, or a specific issue short ID rather than asking for “everything.” If the first pass is too sparse, widen the time window before removing filters. The best sentry usage usually comes from one focused query followed by one refinement, not a single giant search.