Agent configuration

Give AI SRE the context and playbooks that currently live in your team's heads. A custom prompt carries instructions into every conversation in the team, while skills are detailed playbooks the agent loads only when they fit the situation.

Go to AI SRE → Integrations → Customize

Custom prompt

The custom prompt is added to the agent's instructions, so AI SRE follows it in every conversation. Use it for what the agent can't learn from your data:

• How your services relate to each other, and which ones are critical.
• Naming conventions for your sources, dashboards, and log fields.
• Escalation rules and who owns what.
• Output preferences, like summarizing findings before proposing a fix.

Keep it to standing instructions. Anything that matters for only one kind of investigation belongs in a skill, so it doesn't compete for the agent's attention every time.

Skills

A skill is a playbook for one job, like investigating a latency spike in the checkout service or drafting a status page update. AI SRE loads a skill only when it's needed, so a skill can be long and specific without slowing down unrelated conversations.

Click Create skill and write it as a SKILL.md document:

---
name: checkout-latency
description: Investigate a latency spike in the checkout service. Use when checkout p95 latency crosses 500 ms.
---

1. Chart p95 latency from the `checkout-api` source, grouped by endpoint.
2. Compare it against the `payments` and `inventory` upstreams over the same window.
3. Check the last three deploys in GitHub for changes to the checkout path.
4. Summarize the findings on the incident before proposing a fix.

The name and description in the headline are how the agent decides when to reach for the skill. Everything below the headline is the instructions it follows.

Run a skill on demand

Type / in the chat composer and pick a skill to have AI SRE follow it for that message. Use it when you already know which playbook applies and don't want to wait for the agent to find it.