Skip to main content

Context Setup

Interviews you to set up and improve a project's context for incident response: checks the integrations, imports services into the Context Graph, writes skills for the project's agents, and scores how ready it is.

How it works​

  1. Runs when: you start it, on day 0 to set a project up, then again on day 1, in week 1 and whenever your systems change. Each run continues from the readiness report the last one left.
  2. Takes stock: reads the project's connections and checks that each returns data, then its skills, integration context forms, old guides and the Context Graph.
  3. Agrees the domains and sources of truth: proposes which domains the project holds from your teams, CODEOWNERS, paging services and channels, and where the truth about services, owners and dependencies lives (a CMDB, PagerDuty, a service catalog, CODEOWNERS, Kubernetes, a wiki page).
  4. Interviews you: one category at a time (components, alerts, logs, metrics, runtime, code, deploys, data stores, tickets, comms, incident history). It explores each integration first, so you confirm what it found instead of describing it.
  5. You approve: each skill before it is committed, and each Context Graph import before it is written.
  6. You get: tested skills for each integration and component, written for an agent in the middle of an incident; the domain's team, services, owners and dependencies in the Context Graph; and a readiness score out of 100 with the pending work in order and one recommended next step.

What it scores​

AreaPoints
Integrations return data10
Domains and owners10
Service inventory15
Integration skills, with tested queries20
Architecture and component skills15
Code mapped to components10
Alerts map to services10
Incident history, and a proven run on a past incident10

It also says how ready the project is for code review and for questions about your systems.

What it needs​

  • Your code host (GitHub, GitLab, Bitbucket or Azure DevOps): read repositories, CODEOWNERS and deploy workflows
  • Observability (Datadog, Grafana, Prometheus or similar): read logs and metrics to test each query it writes down
  • Alerts and tickets (PagerDuty, Jira, ServiceNow): read services, assignment groups and past incidents
  • Anything else you connect: it uses whatever is there, and gives you a connect link for what is missing