How k9 agents plan features end-to-end
A deep dive into how k9 generates PRDs, ADRs, and technical designs from a single brief.
One of the most common questions we get: "How does kill9 go from a one-line description to a full implementation plan?" This post walks through the planning pipeline.
The input
You type something like: "Add OAuth2 login with Google and GitHub." That's it. One sentence.
What happens next
The orchestrator (k9) routes your request to the planning agent (kpm). kpm generates a Product Requirements Document covering scope, acceptance criteria, and edge cases. It then hands off to the architecture agent (karch) for an Architecture Decision Record — which auth library, session management strategy, token storage.
Quality gates
Before any code is written, kpm presents a questionnaire: "Which auth providers should we support? Should we include SSO/SAML? What session timeout?" Your answers shape the implementation. Nothing proceeds on assumptions.
Stories and delegation
Once the plan is approved, the lead agent (klead) breaks it into stories and delegates to developer agents (kdev) and test agents (ktest). Each story gets its own worktree — isolated branch, zero conflicts.
The result
From one sentence to a planned, architected, story-broken feature — typically in under 2 minutes. The agents don't just plan. They execute.