Introduction

For teams already using Cursor, Claude Code, or similar runtimes who need auditability across sessions — not another autocomplete layer.

After this page you can answer: what fails today, what changes after install, and whether your project matches the adoption profile.

---

The problem

Your agents are strong at coding. They are weak at finishing the SDLC loop:

• context collapses across sessions — the next agent does not inherit operational state
• process steps get skipped — QA, tickets, deploy proof are optional in chat
• “done” is a language claim — no gate checks filesystem or board state before advance

That failure mode is structural. It is not fixed by a larger context window or a smarter model.

---

The consequence

Symptom	Cost
Work lands on main without isolation	review noise, rollback risk
No evidence folder	audits and postmortems lack artifacts
Stale work tracker	planning diverges from repo reality
Cold next session	re-explaining intent burns tokens and time

---

The solution

AI Native SDLC is infrastructure for long-running AI engineering workflows on top of your existing agent runtime.

It adds:

• Deterministic workflow — stages, manifests, resume
• Gates — PASS only when checks succeed
• Memory + handoffs — durable state outside the chat
• Observability — traces and evidence paths you can inspect

Install does not replace Git, CI, or Plane. It coordinates them around agent execution.

npx @jambu/ai-native-sdlc@latest install

Then in the agent:

/sdlc:init

---

See it before the taxonomy

1. Show Me — one intent → tickets → worktrees → evidence → deploy (artifact counts)
2. Home splash — same chain, without vs with, when to install
3. How It Works — install commands and layer map
4. Architecture — internal Procedural Cognitive Runtime Architecture (read after outcomes make sense)

---

When to install

Use when

• Multi-session or multi-agent delivery (> ~1 week horizon)
• Audit, compliance, or regulated environments
• Teams on Cursor / Claude Code / OpenClaw with repo-bound work
• Epics that need parallel worktrees and PR isolation

Do not use when

• One-off scripts or weekend spikes
• Throwaway prototypes under ~500 LOC
• You only need inline completion — no SDLC loop

---

What this documentation is

Section	Purpose
OVERVIEW	Outcomes, install map, workflow vs runtime
ARCHITECTURE	Harness decomposition (after Show Me)
BUILD	Playbooks, workflows, gates
OPERATE	Lifecycle, materialization, distribution
REFERENCE	ADRs, glossary, risks

---

Macro vs micro

Scale	Command	Outcome
Macro	/sdlc:goal	Run manifest, waves, Plane hierarchy, deploy
Micro	/implement-feature	One ticket, one worktree, one PR

Details: Lifecycle.

---

Source of truth

Source	Owns
Plane	Work items, sprint state
.sdlc/INDEX.md	Scoped operational navigation
.sdlc/context/decisions/	ADRs
.sdlc/sdlc.yaml	Runtime configuration

Reference implementation: jambu-ai/ai-native-sdlc.