Skip to main content

Product Management Lifecycle

A business-led roadmap is the blueprint for a successful digital and AI transformation.

AI agents build governed. Humans ship trusted. 80% autonomy, 100% accountability.

Golden Path: Your First Governed Sprint


Phase 1: Discover (30 min)

Who: You (HITL) steer direction. product-owner agent validates business fit.

What: Find problems worth solving. Write your PR/FAQ north-star document.

Why: Most digital transformations fail to capture expected value because they skip discovery. Building without validation equals building the wrong thing. PR/FAQ forces precision.

What-if skip: Pet projects, wasted sprints, no customer alignment. Your team builds features nobody asked for.

How

# Write your north-star document
/product:pr-faq my-product

# Explore the opportunity space (optional, 30 min max)
# Skill:lean-canvas
# Skill:opportunity-solution-tree
# Skill:identify-assumptions

Output

  • pr-faq.md — 6-section north-star (vision, problem, solution, target user, call-to-action, success metrics)
  • opportunity-solution-tree.txt — Outcomes → solutions → experiments
  • assumptions.md — Riskiest 5 bets with validation plan

Phase 2: Specify (1 hour)

Who: product-owner agent drafts spec. You approve scope.

What: Turn validated ideas into precise specifications. Focus on what to build, not how.

Why: Ambiguous specs cause scope creep and rework. Spec-Driven Development makes specs executable artifacts, not disposable documents. Clear ACs prevent agent confusion.

What-if skip: "Just code it" syndrome. Unbounded scope. Agents work on wrong features. No acceptance criteria to validate against.

How

# Generate feature specification from natural language
/speckit.specify "voice-enabled RAG chatbot for cloud operations"

# Clarify underspecified areas (up to 5 targeted questions)
/speckit.clarify

# Review generated INVEST user stories
cat stories.csv

Output

  • spec.md — Problem statement, user personas, success metrics
  • stories.csv — INVEST format (title, AC, dependencies, KR mapping, agent assignment)
  • edge-cases.md — Negative scenarios and error handling requirements

Quality Gate: Every story must have ≥3 acceptance criteria. No story worth >13 points.


Phase 3: Plan (1 hour)

Who: cloud-architect agent designs. product-owner validates alignment.

What: Design technical approach. Map specialist agents to tasks. Validate against ADLC principles.

Why: 80% of transformations fail on scoping, not execution. Planning prevents over-engineering, under-engineering, and agent coordination conflicts.

What-if skip: Over-engineering (Aurora for fewer than 50 users), under-engineering (no auth), agents stepping on each other's work.

How

# Generate implementation plan from spec
/speckit.plan

# Break plan into dependency-ordered tasks
/speckit.tasks

# Review cloud-architect's technical design
cat plan.md

Output

  • plan.md — Technical approach, alternatives considered, trade-offs documented
  • tasks.md — Ordered list with agent assignments and dependencies
  • risks.md — Blockers, escalation paths, constitution checks
  • Agent delegation matrix (who owns which story/task)

Quality Gate: Every task has a single owner. No "TBD" dependencies. Constitution compliance validated.


Phase 4: Build (2–4 hours)

Who: AI agents (python-engineer, infrastructure-engineer, etc.) execute. You review and ship trusted.

What: Execute tasks with specialist agents. PDCA until quality gates pass. Each session produces code, tests, and evidence.

Why: Agents prepare, humans decide. Governed execution with evidence prevents drift and testing theater. No evidence = no completion.

What-if skip: Unreviewed code, no evidence trail, testing theater, NATO violations (completion claims without proof).

How

# Execute implementation according to plan
# Specialist agents handle domain work:
# - python-engineer → CLI, API, business logic
# - infrastructure-engineer → Terraform, ECS, IAM
# - qa-engineer → Test validation, coverage

# Monitor progress
/metrics:daily-standup

# Review evidence in tmp/
ls -la tmp/adlc-framework/test-results/
cat tmp/adlc-framework/coordination-logs/product-owner-*.json

Output

  • Working software on disk (code + tests + docs)
  • Test evidence in tmp/adlc-framework/test-results/
  • PDCA round logs in tmp/adlc-framework/coordination-logs/
  • No story marked "Done" without test execution output

Quality Gate: ≥90% test coverage on critical paths. Zero NATO claims (all completion claims backed by file evidence). All ACs validated with tests.


Phase 5: Measure (30 min)

Who: observability-engineer agent collects metrics. You read the dashboard.

What: Track sprint health, DORA metrics, agent consensus scores. Evidence-based ceremonies.

Why: What gets measured gets managed. Ceremonies create the evidence trail that prevents NATO violations and stakeholder surprises.

What-if skip: No sprint visibility, invisible blockers, "done" claims without proof, board surprises.

How

# Daily visibility into sprint health
/metrics:daily-standup

# Sprint closure with demo + metrics + DORA
/metrics:sprint-review

# View DORA dashboard
open https://adlc.oceansoft.io/docs/dora

Output

  • DORA dashboard (deploy frequency, lead time, change failure rate, MTTR)
  • Sprint velocity (planned vs actual)
  • Blockers and resolved issues log
  • Agent consensus scores (PO, CA, QA)

Quality Gate: Velocity trend charted. No unexplained deltas >20%. All blockers documented with resolution paths.


Phase 6: Iterate (30 min)

Who: Full agent team reflects. You decide what to change.

What: Retrospective with 4L review (Liked, Learned, Lacked, Longed-for). Extract patterns. Improve the process.

Why: The process that builds products must itself improve continuously. Improvement compounds over sprints. Org learning > one-time fixes.

What-if skip: Same mistakes repeated, process stagnation, team burnout, no organizational learning.

How

# Sprint retrospective with improvement actions
/metrics:sprint-retro

# Session retrospective for continuous learning
/speckit.retrospective

# Review process improvements
cat framework/retrospectives/YYYY-MM-DD-sprint-*.md

Output

  • 4L review (Liked, Learned, Lacked, Longed-for)
  • 3–5 improvement actions assigned to next sprint
  • Updated process patterns in .claude/
  • Constitution update (if governance gaps found)

Quality Gate: Every sprint improvement tracked. Metrics show velocity trend ≥0% (no degradation). Retrospective consensus ≥70%.


LEAN/5S Applied to Product Management

PrincipleApplicationEvidence
Sort42 PM skills curated from 65+ (removed generic, overlapping, non-PM).claude/skills/product/ directory — 32 items deleted
Set in Order4 phases: Discovery (16 skills) > Strategy (14) > Growth (4) > Utility (8)Organized by product lifecycle stage
ShineQuality gates at each phase transition (INVEST scoring, constitution check)PO+CA+QA sequential scoring after each deliverable
StandardizeThis golden path = consistent process across all digital productsSame 6 phases for CloudOps, xOps, DevOps, Agentic-AI
Sustain/metrics:sprint-retro keeps the process improving every sprintRetrospective evidence in framework/retrospectives/

By Persona

Solo Startup Founder

You have: An idea and AI agents. No team yet.

Your path (4 hours to first governed sprint):

/product:pr-faq my-product          # 30 min — north-star document
/speckit.specify "my product description" # 1 hour — precise spec with user stories
/speckit.plan # 1 hour — technical design
# Agents execute (2–4 hours)
/metrics:sprint-review # 30 min — evidence-based completion

Value: First governed sprint with zero team. AI agents do 80% of the work, you make 100% of the decisions. Evidence trail from day 1. Board-ready demo.

CxO / Enterprise Leader

You need: A digital transformation roadmap with board-ready evidence.

Your path (1 day to board-ready roadmap):

/product:pr-faq                    # Vision statement that inspires the team
# Skill:lean-canvas # Business model hypothesis on one page
/speckit.plan # resource planning with agent delegation
/metrics:sprint-review # Evidence-based progress for your board

Value: Board-ready roadmap with measurable KPIs, not PowerPoint promises. DORA metrics prove execution velocity. Quarterly demos with evidence.

Product Owner

You manage: Sprints with AI agents. Daily ceremony cadence.

Your path (ongoing sprint management):

/speckit.specify + /speckit.tasks          # Spec to task breakdown
/metrics:daily-standup # Daily visibility (5 min read)
/metrics:sprint-retro # Continuous improvement

Value: 42 PM skills organized by phase. Ceremony automation reduces overhead from 8h/week to under 2h. PDCA quality gates. Agent consensus scoring (PO+CA+QA).


Reality Check: xOps Pilot Evidence

The xOps Command Centre is the live proof point for this golden path. Real execution evidence:

PhaseEvidenceResult
DiscoverF2T2EA north-star + 13 INVEST storiesPO consensus 84%, CA consensus 82%
Specify6-phase operational cycle with personas7 components, 6-layer architecture
PlanTechnical design + $110–460/mo cost modelAccepted by HITL + CA
Build1,430 LOC across 10 modules98.3% cross-validation accuracy
MeasureDORA dashboard (3/4 GREEN)100% BV delivered (1,472/1,472 points)
IterateRetrospective actions → next sprint4-week cycle time, improving

Not theoretical. Real. Running. Now.


PM Skills Reference

All skills are in .claude/skills/product/ organized by phase:

PhaseCountKey Skills
Discovery16lean-canvas, opportunity-solution-tree, identify-assumptions, customer-journey-map, interview-script, user-personas, job-stories
Strategy14product-vision, product-strategy, competitive-battlecard, north-star-metric, market-sizing, pricing-strategy, swot-analysis
Growth4growth-loops, ab-test-analysis, cohort-analysis, sentiment-analysis
Utility8pre-mortem, value-proposition, startup-canvas, beachhead-segment, analyze-feature-requests

Each skill is copy-paste ready with examples.


Common Mistakes (Anti-Patterns)

MistakeWhy It FailsFix
Skip Discover, jump to BuildNo validation. Build wrong thing. Sprint failed before code started.Always start with PR/FAQ. Non-negotiable.
Spec with 50+ stories in first sprintScope creep. Agents confused. Nothing finished.Max 5 stories per sprint. Start small, iterate.
No acceptance criteria in specAgents guess what "done" means. Scope expands.Every story must have ≥3 ACs. Template enforced.
Skip Plan phaseArchitectural debt. Over-engineering. Agent coordination chaos.Cloud-architect approval required. Takes 1 hour.
Build without testsNATO violations (completion claims without proof).Tests before every "done" claim. Non-negotiable.
No ceremonies/metricsFlying blind. Board surprises. No org learning.Daily standup + sprint review + retro. Automated.

Success Metrics by Role

HITL (You)

  • Sprint velocity trend ≥0% (no degradation)
  • Board Q&A answerable without rework (DORA data ready)
  • Agent consensus ≥70% (no disputed completions)
  • Retrospective actions tracked to next sprint

Product Owner (Agent)

  • Spec clarity score ≥85% (agent questions < 3)
  • Story size consistency (no outliers >13 pts)
  • INVEST compliance 100% (all stories auditable)

Specialist Agents

  • Test coverage ≥90% critical paths
  • Zero NATO violations (all claims backed by evidence)
  • Task dependency compliance (no surprises)

Org

  • Cycle time improving (baseline → -10% per sprint target)
  • Rework rate under 10% (spec quality high)
  • On-time delivery ≥95% (planning accurate)

Quick Reference: Command Cheat Sheet

# Discover
/product:pr-faq "product name"
# Skill:lean-canvas
# Skill:opportunity-solution-tree

# Specify
/speckit.specify "product description"
/speckit.clarify # up to 5 Q&A rounds
/speckit.checklist # generate quality checklist from spec

# Plan
/speckit.plan
/speckit.tasks
/speckit.analyze # cross-artifact consistency check

# Build
/metrics:daily-standup # view sprint progress
# (agents execute /speckit.implement)

# Measure
/metrics:sprint-review # sprint closure + DORA
/metrics:update-dora # refresh DORA dashboard

# Iterate
/metrics:sprint-retro # 4L review + actions
/speckit.retrospective # session learning

# Always
git diff --stat # verify changes on disk
ls tmp/adlc-framework/test-results/ # check evidence

Integration with Framework

This golden path is part of the larger ADLC framework:

  • Constitution: 7 non-negotiable principles (Acceptable Agency, Interoperability, Evaluation-First, etc.)
  • Governance: 58 checkpoints ensure 99.5% compliance
  • Commands: 80+ slash commands automate each phase
  • Skills: 414 curated components in the marketplace
  • MCPs: 24 external integrations (AWS, GitHub, Jira, Azure, Slack, etc.)

Start here. Integrate deeper as you scale.


Last Updated: March 2026 | Status: Proven golden path (xOps pilot validation) | Maintenance: Retrospective-driven improvements every sprint