# Incident Triage

First-pass triage of a possible production incident for Paradev's hosted services.

- Repository: https://github.com/ParadevAI/Paradev (15823 stars)
- Category: Engineering
- License: see repository
- Last updated: 2026-11-01
- Runs scripts: no, instructions only

## Install

Claude Code: `npx skills add ParadevAI/Paradev --skill incident-triage -g -a claude-code -y`
Codex: `npx skills add ParadevAI/Paradev --skill incident-triage -g -a codex -y`
Cursor: `npx skills add ParadevAI/Paradev --skill incident-triage -g -a cursor -y`
Gemini CLI: `npx skills add ParadevAI/Paradev --skill incident-triage -g -a gemini-cli -y`
Claude Code plugin marketplace: `/plugin marketplace add ParadevAI/Paradev`

## Description

First-pass triage of a possible production incident for Paradev's hosted services. Gathers signals (recent deploys, Sentry spikes, health checks), proposes a severity and a status.paradev.com message, then stops for human approval. Use when someone reports an outage, errors spike, or asks "is something down?".

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Incident triage

You gather facts and draft. A human decides and posts. The runbook is
`docs/runbooks/incidents.md`; read its "Severity" and "Status page updates" sections before you
start.

## Rules

- Read-only. Do not deploy, roll back, disable workflows, change flags, resolve Sentry issues,
  or run SQL that writes.
- Do not post to status.paradev.com, Slack, Discord, Plain, X, or any other channel. Do not declare
  an incident. Your output is a draft in this conversation.
- If a tool is not connected or a check fails, write "not checked" and the reason. Do not guess.
- Use UTC times everywhere.

## 1. Frame the report

Write one line each:

- What users see, in their words.
- When it started (first report, or first signal you find).
- Who is affected: everyone, one region, one feature, one version, or one user.

If only one user or one machine is affected, stop here. Tell the human that this is a support
issue, and suggest `superset doctor`.

## 2. Gather signals

Run the checks that can run in parallel at the same time. Record each result with its time.

The rest: https://github.com/ParadevAI/Paradev/blob/HEAD/.agents/skills/incident-triage/SKILL.md