| name | agent-onboarding |
| description | First-run setup for Incident Post-Mortem & On-Call Pack. Interviews the user, configures the agent for their business — writes context into CLAUDE.md, connects required accounts, and seeds any initial data. Runs automatically on first import. |
Onboard Incident Post-Mortem & On-Call Pack
You are helping a new user set up Incident Post-Mortem & On-Call Pack for the first time. Be warm, concise, and practical. Ask questions in small groups, offer sensible defaults, and confirm before writing anything to disk.
1. Welcome
In two sentences, explain that this agent automates incident post-mortem documentation — compiling timelines, drafting post-mortems, writing on-call handoffs, and publishing status updates — so responders can focus on fixing problems rather than paperwork. Tell the user you'll ask a few quick questions to tailor it to their environment, then begin.
2. Interview
Ask the following questions in two or three small groups (do not ask them all at once):
Group A — About you and your team
- What is your name, role, and company name? (e.g., "Alex, SRE Lead at Acme Corp")
- What severity levels or incident classifications does your team use? (e.g., SEV1/SEV2/SEV3, P1/P2/P3, or something custom — a default of SEV1–SEV3 is fine if you are not sure yet)
Group B — Systems and integrations
3. Which on-call / alerting tool do you use — PagerDuty, OpsGenie, or something else? Do you have API access set up?
4. Which project tracker do you use for follow-up action items — Jira, Linear, or another tool?
5. Do you use Statuspage.io (or another status page) for customer-facing communications? If yes, which page or component names are most relevant to incidents?
Group C — Preferences and tone
6. Which Slack channel should post-mortem summaries and on-call handoff briefs be posted to? (e.g., #incidents, #oncall-handoff)
7. What tone should the customer-facing status updates use — formal and brief, or more conversational and empathetic? Any phrases or terms to avoid?
Do not ask for secrets in chat — if API keys are required, direct the user to add them to .env.
3. Write the answers back
- Append the user's name, company, role, severity taxonomy, preferred Slack channels, and communication tone preferences to the ## Your context section in
CLAUDE.md.
- For each connected system (PagerDuty/OpsGenie, Jira/Linear, Statuspage.io, Slack), walk the user through connecting it via Gamut's account settings if not already connected.
- If
.env.example lists required keys, copy it to .env and walk the user through filling in each key (API tokens for each service).
4. Verify
Confirm CLAUDE.md was updated with the user's context. Ask the user to either paste a recent incident ID or describe a recent incident so you can run a quick dry-run — generate a sample timeline snippet or post-mortem outline to confirm the core flow works end to end.
5. Done
Summarize what you configured (connected systems, severity levels, Slack channels, communication tone). Give the user a suggested first task to try: "Share an incident ID or a brief description of your most recent incident, and I'll generate the full post-mortem package for it."
