Preface¶

When an online alert pops up, developers typically do the following: first open Sentry to check the Issue list, then switch back to the terminal or IDE to check logs, trace code, and draft fixes. Switching back and forth between the Dashboard and the conversation window easily breaks the context; if you just ask an AI “What unresolved errors are there in production recently?”, it usually cannot access the real-time data in your Sentry project and can only give general suggestions.

sentry is an Agent Skill maintained by OpenAI in the .curated directory of the openai/skills repository. It integrates the read-only query capabilities of the official Sentry CLI into the Skill instructions, allowing tools that support Agent Skills such as Cursor and Codex CLI to call the sentry command in a standardized way when users request actions like “checking Issues, summarizing production errors, or pulling basic health data” instead of guessing randomly.

The following explanation follows the order of “what it is → what it can do → how to install it → how to use it → what to note”. The commands and parameters mentioned in the article have been cross-verified from the official SKILL.md and the Sentry CLI documentation.

What it is¶

The positioning of the sentry Skill can be summarized as: Perform read-only observability queries via the Sentry CLI.

• Publisher: The Skill instructions are maintained by OpenAI and included in skills/.curated/sentry; it relies on the official Sentry sentry CLI (cli.sentry.dev) as its underlying dependency.
• Trigger scenarios (written in the description field of SKILL.md): Activated when the user requests checking Sentry Issues or Events, summarizing recent production errors, or pulling basic health data.
• Capability boundaries: Mainly focused on read-only queries; authentication, organization/project identification, pagination, and retries are handled by the CLI, while the Agent focuses on organizing queries and presenting results.

The Skill follows the universal SKILL.md format (Agent Skills Open Standard), and the same folder structure can be reused across different AI programming tools.

Core Functions and Highlights¶

1. Complete read-only workflow around Issues¶

The Skill organizes commonly used CLI subcommands into a fixed workflow, mainly including:

Issues use short IDs (such as ABC-123), not numeric IDs. List queries support Sentry search syntax, for example is:unresolved environment:production.

2. Automatic org / project identification¶

The CLI will automatically infer the target project from the DSN in .env, source code, default configuration values, directory names, etc.; if inference fails, you can explicitly pass in {your-org}/{your-project}. This is very convenient for monorepos or multi-service projects.

3. Agent-oriented JSON output convention¶

For programmatic processing, the Skill requires using --json; when you need to simplify fields, combine it with --json --fields. For uncovered APIs, you can use sentry api, and use sentry schema to discover available endpoints.

4. Built-in security and presentation specifications¶

The Skill clearly requires:
- When not authenticated, guide the user to run sentry auth login locally, or set the SENTRY_AUTH_TOKEN environment variable; do not paste the full Token in the conversation.
- Desensitize PII (email, IP, etc.) in the output, do not print the original full stack trace; do not echo authentication credentials.
- Clearly state when there are no results, and the Issue list should display key fields such as title, short_id, status, time, and count.

5. Default query parameters¶

The default values agreed by the Skill (can be overwritten in the command):
- Time range: 24h (--period 24h)
- Environment: Production (write environment:production in the query)
- Maximum number of entries: 20 (--limit 20)

Installation and Enablement¶

The Skill itself is a directory (at minimum containing SKILL.md). Before using it, you also need to install the Sentry CLI on your local machine and complete authentication.

Step 1: Install the Sentry CLI¶

Official installation method (from the original Skill text):

curl https://cli.sentry.dev/install -fsS | bash

Authentication and verification:

sentry auth login
sentry auth status

You can also complete authentication via the environment variable SENTRY_AUTH_TOKEN. The Token should only be configured locally and never sent to the Agent.

Step 2: Install the sentry Skill¶

Codex CLI (verified via official documentation): Use $skill-installer in Codex conversations to install curated skills, for example:

$skill-installer sentry

Or specify the GitHub directory URL:

$skill-installer install https://github.com/openai/skills/tree/main/skills/.curated/sentry

Restart Codex after installation to load the new Skill. OpenAI documentation also states that you can place the Skill directory in the .agents/skills path under the repository or user directory for Codex to scan and discover.

Cursor: Copy the skills/.curated/sentry directory to the project’s .cursor/skills/sentry/ (or user-level Skill directory), just make sure it contains SKILL.md. Cursor will match tasks based on the Skill’s name and description in the Agent context.

Note: The README of the openai/skills repository marks that the repository has been deprecated, and subsequent Codex plugin examples will be moved to OpenAI Plugins; but the content of the sentry SKILL.md and the Sentry CLI commands can still be used directly. Please refer to the current documentation of each tool for the installation method.

Typical Usage Examples¶

List unresolved production Issues from the last 24 hours¶

Standard example from the Skill (consistent with the issue list documentation):

sentry issue list \
  --query "is:unresolved environment:production" \
  --period 24h \
  --limit 20 \
  --json --fields shortId,title,priority,level,status

Explicitly specify org/project when automatic identification fails:

sentry issue list my-org/my-project \
  --query "is:unresolved environment:production" \
  --period 24h \
  --limit 20 \
  --json

View Issue details and Events¶

sentry issue view ABC-123 --json
sentry issue events ABC-123 --limit 20 --json

Use Sentry’s built-in AI for root cause and repair suggestions¶

sentry issue explain ABC-123
sentry issue plan ABC-123

These two commands depend on Sentry-side capabilities and require your account and project permissions to meet the CLI requirements.

Prompt in Agent conversations¶

Example from the Skill’s built-in Golden test:

List the top 10 open issues for prod in the last 24h.

You can use a similar expression in Chinese scenarios: “List the top 10 unresolved Sentry Issues in the production environment in the last 24 hours, sorted by recent activity.” After enabling the sentry Skill, the Agent should perform CLI queries and summarize results in a structured way instead of fabricating an Issue list.

Applicable Scenarios and Notes¶

Who it is suitable for
- Developers, SREs, and DevOps who need to embed online fault troubleshooting into their AI workflow.
- Teams that have integrated Sentry into their projects and want Agents to read-only pull Issues/Events to assist with on-duty shifts, post-incident reviews, and writing incident summaries.
- Teams that want to unify the workflow of “check Sentry → explain → get repair directions” and reduce the mental burden of manually writing CLI commands every time.

Usage Notes
1. Read-only boundary: The Skill description emphasizes read-only queries; write operations such as modifying Issue status or releasing versions are not within the core workflow of the Skill. Do not mistakenly assume that the Agent will automatically modify Sentry configurations.
2. CLI must be ready first: The Skill will not replace you to install the CLI; if it is not installed, you should provide the installation and authentication steps according to the Quick start.
3. Permissions and data security: Production data may contain sensitive information, and the Skill requires desensitization and restrained output; if the team policy prohibits Agents from accessing Sentry, you should disable this Skill at the tool layer or revoke the Token.
4. Default values can be overwritten: The default 24h / production / limit 20 is suitable for quick inspections; you need to adjust them in --query and --period for cross-environment or longer cycles.
5. Repository status: The curated skills are still hosted in openai/skills; for long-term distribution, please follow the updates of OpenAI Plugins and the Codex Skills documentation.

Summary¶

The value of the sentry Skill lies in encapsulating the Sentry CLI’s Issue query, Event viewing, AI explanation, and repair plan into an instruction set that Agents can execute repeatedly. When an online problem occurs, you no longer need to copy and paste back and forth between the Dashboard and the chat box — just configure the CLI and Token locally, and let the Agent query, summarize, and desensitize and present results according to the Skill specifications.

Official Skill directory: https://github.com/openai/skills/tree/main/skills/.curated/sentry
Sentry CLI documentation: https://cli.sentry.dev/