Lifecycle

From intent to deployed product

The lifecycle is a sequence of workflows composed by commands. At macro scale, /sdlc:e2e runs the full chain. At micro scale, /implement-feature runs one vertical slice.

Invariants: work-items ticket required, worktree isolation, deterministic gates, handoff on completion. The ticket system is whichever integration occupies the work_items slot — not a fixed vendor.

---

Phase 0: Install and init

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

Init discovers the repository, binds workspace.target_root, detects profile, generates integration manifests, and sets status: operational. After init, the full lifecycle is available on this project.

See Distribution.

---

Phase 1: Context hydration

Trigger: session start (runtime hook) or /hydrate-context

Reads: INDEX (scoped), LATEST handoff, sdlc.yaml, adapter memories, active run if any.

Output: agent resumes without full repository scan.

/sdlc:discovery produces strategic repo analysis before intent classification — distinct from E2E visual discovery stage.

---

Phase 2: Specification

Trigger: /spec:create {ticket-id}

Workflow: spec-flow.yaml → spec-writer agent

Actions:

1. Read ticket from active work-items integration
2. Produce spec in .sdlc/specs/
3. Enrich ticket with technical notes (implementation plan)
4. Register spec in INDEX
5. Create git worktree

The spec’s Technical Notes section IS the implementation plan — vertical, self-contained.

---

Phase 3: Feature implementation

Trigger: /implement-feature (inside worktree)

Workflow: feature-flow.yaml

context-hydration → planning → implementation → review → qa
    → pr-creation → work-items-in-review → work-items-closure

Planning

Planner produces affected files, interfaces, atomic tasks, E2E scenarios. Optional ADR draft if architectural decision required.

Implementation

Code following adapter rules and memories. Build and type checks must pass for the active profile.

Review

Reviewer validates against rules and ADRs. APPROVED | CHANGES_REQUESTED | BLOCKED.

QA

Browser-native validation via user-configured MCP (typically Playwright). Evidence to .sdlc/evidence/{TICKET}/ in worktree. Worktree gate blocks QA in main workspace.

PR and ticket sync

PR opened with QA summary. Ticket moves to In Review immediately. On merge: Done, comment posted, evidence uploaded per integration API.

---

Phase 4: Deployment

Trigger: /deploy-release

Workflow: deployment-flow.yaml → deployer agent

Staged pipeline: build → version bump → staging → staging smoke → production → post-deploy validation.

Follow-up: /sdlc:post-deployment — version audit, smoke with video, observability error check, evidence manifest, rollback gate.

Deploy target comes from active deploy slot integration — generated at init, not bundled.

---

Phase 5: Observation and remediation

Trigger: /fix-incident or CI incident workflow

Workflow: incident-flow.yaml → incident-resolver

Observability integration (if configured) provides error context. Agent reproduces, writes regression test, opens fix PR in hotfix worktree.

---

Macro path: /sdlc:e2e

hydrate → classify → discovery → decomposition → work-items-hierarchy
    → architecture → workspace-rebind → wave-planning
    → execution (per ticket, parallel by wave)
    → integration → deploy → post-deploy → board-sync → reflect → handoff

Each ticket in execution runs full feature-flow via subagent dispatch in its worktree. Orchestrator never implements {target_root}/** in main session.

Per-phase vertical trace

Phase	Trigger	Data	Transform	Output	Observability
Intent	/sdlc:e2e	user intent string	manifest init	manifest.yaml	trace event
Hydrate	stage-start	INDEX, LATEST, sdlc.yaml	sessionStart hook	context packet	events.jsonl
Execution	wave ticket	spec, worktree	feature-flow subagent	PR, evidence	manifest ticket fields
Deploy	stage-complete deploy	CI, staging URL	deploy-release	production URL	post-deploy manifest
Close	mark-done	evidence dir	board-sync, reflect	Plane Done, handoff	QA manifest

---

Closure requirements

Requirement	Enforced by
Feature-flow success criteria	sdlc_gate_check.py
Evidence uploaded (if policy enabled)	work-items-closure stage
Non-stub handoff	mark-done gate
SDLC state committed	mark-done gate
/sdlc:reflect pass	reflect stage gates
Doctor operational pass	/sdlc:doctor

---

Lifecycle slot coverage

Each phase binds to an integration slot at init. Examples:

Phase	Slot	Example providers
Tickets	work_items	Plane, Jira, Linear, Notion
Source + CI	source_control, ci	GitHub, GitLab
QA	stack declaration	Playwright, Cypress
Observability	observability	PostHog, Sentry, Datadog
Deploy	deploy	Cloudflare Pages, Railway, Fly.io

Same workflow YAML regardless of provider. See Integrations.

---

Cross-cutting: dependency gate

When the active profile uses a package manifest (package.json, etc.), lockfile sync runs before commit. Enforcement: git hooks + adapter rule. Platform-conditional deps must be explicit for CI reproducibility.