Agentic PM Playbook

The open operating system for modern product teams

Evidence in.
Outcomes out.

A practical playbook for replacing opinion-driven roadmaps with continuous discovery—and giving every product team a disciplined crew of AI collaborators.

Explore the playbook Browse the curriculum
7 learning modules6 specialist agents1 closed-loop system
OUTCOMES OVER OUTPUT✦CONTINUOUS DISCOVERY✦SMALLEST TEST FIRST✦JUDGMENT STAYS HUMAN

The operating loop 01

From raw signal to shipped learning.

The playbook connects discovery and delivery in one traceable loop. Every decision carries its evidence, assumptions, and intended outcome forward.

01

Sense

Turn scattered feedback into usable evidence.

Agents collect interviews, support tickets, and behavioral data—then cluster the signal without sanding off contradiction or uncertainty.

ProducesEvidence-tagged signal clusters
INPUT STREAMSInterviewsSupportUsage
Clusterthemes + tensions
Desired outcomeOpportunity Solution Tree
The system never stops listeningChoose each stage to trace the loop

The stack 02

Three layers.
One coherent practice.

01 / HUMAN

Judgment

People own truth, priority, ethics, and the call to proceed—or kill.

What matters?
02 / SYSTEM

Shared context

A living tree connects outcomes to opportunities, solutions, assumptions, and evidence.

What do we know?
03 / AGENTS

Production power

Specialists synthesize, structure, design, test, review, and ship—with explicit handoffs.

What can we accelerate?

The agent team / 03

A crew, not a chatbot.

Six specialized agents installed by setup.sh. Delegate explicitly, or let Claude route automatically.

pm

Discovery, OST maintenance, signal synthesis, user stories

architect

System design, ADRs, schema review

engineer

Implementation, refactoring, debugging

qa

Test strategy, test writing, edge case hunting

reviewer

Correctness, security, and performance audits

release-manager

Triage and merge open PRs, then ship

See the human–agent–Compass workflow

The curriculum / 04

Learn the practice by running it.

The curriculum teaches the operating model first (why), then the tools (how), using a shared fictional product so you can practice before running it on your own.

Module Title Time
0 The Operating Model Half day
1 Environment Setup Half day
2 Your First OST 1 day
3 Signal Synthesis 1 day
4 Experiments & Investment 1 day
5 The Agent Team 1 day
6 Cadences & Health Half day
Capstone One Full Cycle on Your Real Product 1–2 weeks
Full curriculum index with prerequisites →

The capability library / 06

Twenty-one skills, one connected operating system.

Discovery, planning, delivery, human review, integrations, scheduled workflows, and engineering discipline—installed by setup.sh.

PM skills
pm-setup One-time config — writes pm-config.md for your project
pm-coach Thinking partner for discovery, OST review, experiment design
ost-workflow Build, extend, and health-check an Opportunity Solution Tree
pm-signal-synthesis Turn interviews, tickets, and reviews into OST-ready clusters
experiment-workflow Design, run, and close assumption tests
investment-gate Assess readiness against the Progressive Investment ladder
integration-routing Route product-state capabilities through configured providers
human-review-workflow Create and apply asynchronous human review decisions
okr-workflow Create, review, and connect OKRs to desired outcomes
roadmap-workflow Manage evidence-backed Later, Next, and Now horizons
status-report-workflow Generate cross-provider product status reports
jira-workflow Create and update Jira issues from discovery artifacts
compass-workflow Maintain Compass discovery and product state
compass-feedback-triage Process open feedback into real product actions
compass-resolver Autonomously deliver an approved Compass roadmap item
delivery-completion-watcher Reconcile reviewed delivery work after CI or deployment changes
agentic-pm Full-cycle PM workflow orchestration
release-manager Triage, merge, and ship open PRs
Engineering discipline skills
design-before-code Pre-implementation design gate — explore context, propose approaches, get approval before any code
test-first TDD iron law — RED test required before any production code, with rationalization counters
verify-done Verification gate — run the command, read the output, cite evidence before claiming done

Quick start

Prerequisite: Claude Code must be installed — run claude --version to confirm.

  1. 1

    Clone the repo

    Get the playbook onto your machine.

    git clone https://github.com/rbcodelabs/agent-pm-playbook cd agent-pm-playbook
  2. 2

    Preview the install

    See exactly what setup.sh will symlink into ~/.claude/ before it does anything.

    ./setup.sh --dry-run
  3. 3

    Install agents + skills

    Symlinks the 6 agents and 21 skills into ~/.claude/ — repo updates propagate automatically.

    ./setup.sh
  4. 4

    Restart Claude Code

    Exit the session and start a new one so the new agents and skills are picked up.

    exit claude
  5. 5

    Verify your install

    Confirm agents and skills loaded by referencing one of each by name.

    "Have the pm agent review my OST" "Run the pm-setup skill"

If you get "unknown skill" or "unknown agent" errors, re-run ./setup.sh --force and restart again.

Operating principles 05

The rules that keep speed honest.

01

Outcomes over output

Start with the measurable change you want—not the feature you want to ship.

02

Continuous discovery

Keep customer contact and evidence flowing every week, not only before planning.

03

Smallest test first

Spend the least needed to challenge the assumption most likely to sink the idea.

04

Judgment stays human

Agents accelerate the work. People remain accountable for truth, ethics, and priority.

Want help adopting this with your team?

RB Code Labs offers facilitated workshops to help product teams adopt this operating model — live, with your real product and real signals. If you want the curriculum accelerated and embedded in your team rather than self-served, get in touch.

Email rick@rbcodelabs.com Visit rbcodelabs.com