Rootly agent skills

Claude Code Cursor Codex Devin

Rootly's plugin for AI coding agents: Claude Code, Codex, Cursor, and Devin. This repository is the canonical source for the skills and generated platform adapters. Marketplace listings are separate distribution channels. It connects your agent to your Rootly account and teaches it how to work an incident, so you can ask "what is going on with checkout right now?", "is it safe to deploy this?", or "who is on call for payments?" and get answers grounded in your real incidents, alerts, services, and schedules.

Connection and server capabilities are documented in the Rootly MCP server repository.

What's inside

Seven skills. Each is a SKILL.md workflow with its own reference material: which Rootly records to pull, how to weigh evidence, what the answer should look like, and a worked example. They cover incident response, alert triage, deploy checks, service health, on-call, incident communications, and retrospectives.
Rootly MCP connection. The plugin bundles Rootly's hosted MCP server, so after a one-time OAuth sign-in your agent reads live incidents, alerts, services, schedules, and retrospectives. No API keys to manage.
Slash commands. Short aliases such as /rootly:respond, /rootly:oncall, and /rootly:retro route to the right skill in Claude Code and Cursor.
Read-only guardrails. The bundled connection requests server read-only mode, with additional read-tool allowlists for Claude Code and Codex. See safety for the boundaries and verification steps.

How it helps

During an incident: impact, recent changes, related incidents, and mitigation options with the evidence behind each, plus who to pull in.
Before a deploy: GO / CAUTION / NO-GO from incident history, service health, and current on-call coverage.
On call: coverage lookups, complete shift handoffs, and swap options.
Afterwards: stakeholder updates drafted for the right audience, and retrospective evidence with facilitation questions.

Skills

Skill	What it does
Incident response	Investigate impact and recent changes; recommend mitigation and responders
Alert triage	Recommend a new incident, attachment, wait, or dismissal — with evidence
Deploy check	GO / CAUTION / NO-GO with reasons; staffing warnings kept separate
Service health	Errors, latency, and traffic against baselines, plus ownership context
On-call	Coverage lookups, complete shift handoffs, provisional swaps
Incident comms	Audience-specific drafts — no invented impact or recovery estimates
Retrospective	Evidence and facilitation questions; humans own the conclusions

Install

Each client installs one plugin named rootly that bundles the seven skills, slash-command aliases, and the Rootly MCP server. Pick your client:

Claude Code

claude plugin marketplace add rootlyhq/agent-skills
claude plugin install rootly@rootly-agent-skills --scope local

Codex

codex plugin marketplace add rootlyhq/agent-skills
codex plugin add rootly@rootly-agent-skills

Cursor

Use a local checkout of this repository as described in installation. The Cursor Marketplace listing is pending; the repository includes both Agent Plugins and Cursor manifests.

Devin

Works in the Devin CLI, Desktop, and cloud sessions.

devin plugins install rootlyhq/agent-skills

Skills only, any agent

npx skills add rootlyhq/agent-skills

This installs the skills as plain SKILL.md files without the bundled MCP server or aliases; connect Rootly MCP through your client yourself.

Then invoke a skill, for example /rootly:incident-response in Claude Code and Devin or $rootly:incident-response in Codex. The first call prompts for Rootly sign-in through the client's MCP flow; never paste a token into chat. Local checkouts, read-only restrictions, and per-client quirks are in installation.

Read-only by design

The skills recommend and draft only — they never page responders, modify incidents, change schedules, publish messages, or act on production. Prompt instructions are not access controls: apply the client restrictions in adapters/ before using live data. See safety and migration.

Rootly's Privacy Policy describes how Rootly handles personal data when you use its services.

Status

1.1 — unreleased consolidation update following the transfer to rootlyhq/agent-skills. The published release is 1.0. Marketplace updates are pending. Read-only by design: the earlier plugins' write actions are not included; see migration.

rootlyhq/rootly