Module 1: Environment Setup

Install the agents and skills, run pm-setup, and stand up the required core plus the recommended notes stack.

Time: Half day Coding required: A little — you’ll run a couple of shell commands. No programming. Prerequisite: Module 0. Do the operating-model module first; this one only makes sense once you know what the tools are for. You will produce: A working core environment — the agents and skills installed, a pm-config.md, and a PM thread that responds — plus, optionally, the recommended notes/discovery stack.

Required core vs. recommended stack

There are two layers here, and conflating them is what makes setup feel heavy. Keep them separate:

The required core is just Claude Code + the playbook’s agents and skills + a pm-config.md. That’s the engine. With only this, you can do every exercise in this curriculum.
The recommended stack (Obsidian, an issue tracker, vault bridges) is where the operating model becomes sustainable day-to-day. It’s worth setting up — but it’s optional, and a solo or non-technical PM with no issue tracker can skip parts of it and still complete the course.

Piece	Layer	What it is	Which shift it serves
Claude Code	Required	The host you run the agents and skills in	The whole model — this is the engine
The agents + PM skills (via `setup.sh`)	Required	6 agents (pm, architect, engineer, qa, reviewer, release-manager) + PM skills (`pm-coach`, `pm-setup`, `ost-workflow`, `pm-signal-synthesis`, `investment-gate`, …)	PM as judgment holder — agents produce artifacts, you decide
`pm-config.md` (via `pm-setup`)	Required	Your config: notes system, tracker (or “none”), current desired outcome	Everything — every skill reads this
Obsidian + Claude Threads	Recommended	A notes home with Claude in the loop	Continuous discovery — signal and synthesis in one place
Issue-tracker integration (Linear / Jira / JPD)	Optional	Work items sync between tracker and notes	Outcomes — work items stay traceable. Skip if you track work in plain notes.
Vault Bridges	Optional	A live link between your product’s repo/docs folder and your notes	Continuous discovery — product context one search away

This module is also the install smoke test. You are the canary for every future adopter. Anywhere you hit friction — a command that fails, a name that’s wrong, a step that assumes context you don’t have — write it down (what you did, what you expected, what happened). That friction log is a real deliverable; it’s what hardens setup.sh and the docs for the public path. Don’t push through silently.

Learning objectives

By the end of this module you will be able to:

Install the playbook’s agents and skills and confirm they’re active.
Run pm-setup and explain what each field in pm-config.md controls.
Get the pm-coach skill to respond as a thinking partner in a thread.
Decide which optional stack pieces you actually need (and skip the rest without guilt).
Capture an install-friction log honestly enough that someone else could fix what you hit.

Concept reading

The one idea worth holding: pm-config.md is the spine. Every PM skill and agent reads it to learn your notes system, your issue tracker (or that you have none), and your current desired outcome. If a skill ever behaves like it doesn’t know your context, the first thing to check is whether pm-config.md exists and is current. You generate it once per project with pm-setup.

Go deeper: README for the project overview, PM Tool Integration Guide for tracker-specific setup (Linear, Jira, JPD, or Markdown-only), and How to Use the Agents for what each agent does.

Hands-on exercise: stand up your environment

Do the Required core (Steps 1–4) first and confirm it works. Then do as much of the Recommended stack (Step 5) as fits how you actually work. Don’t skip the Verify line under each step — that line is the smoke test.

Required core

Step 0 — Install Claude Code (skip if you already have it)

Claude Code is a separate CLI application — different from claude.ai. If you haven’t installed it yet:

Follow the install guide at docs.anthropic.com/claude-code — it has OS-specific instructions.
You also need git. Run git --version in your terminal to confirm it’s there; if not, install it from git-scm.com.

Verify: claude --version returns a version number. If you get “command not found,” complete the Claude Code install before continuing.

Step 1 — Install the agents and skills

The reliable path is to clone the repo and run setup.sh, which symlinks the agents and skills into ~/.claude/ (symlinks mean repo updates propagate automatically):

git clone https://github.com/rbcodelabs/agent-pm-playbook
cd agent-pm-playbook
./setup.sh --dry-run   # preview exactly what it will symlink
./setup.sh             # do it

setup.sh prints a ✓ for each agent and skill it links, then a summary and Restart Claude Code to pick up the new agents. Restart Claude Code afterward. (To restart: type exit in the session and run claude again — or close and reopen the desktop app if you’re using one.)

Optional alternative — the plugin marketplace. If you’d rather install via Claude Code’s plugin system instead of setup.sh, the repo ships a marketplace manifest. ⚠️ Confirm before relying on this: the exact claude plugins subcommand syntax changes between Claude Code versions, and this repo’s plugin manifest is currently a subset of what setup.sh installs. Until that’s reconciled, setup.sh is the install that gives you the full set (all 6 agents + all PM skills). Treat the marketplace path as advanced/optional and verify it against claude plugins --help on your version.

Verify: setup.sh finished with no red errors and a summary listing the agents and skills. Restart Claude Code, then in a session confirm you can reference the pm agent and the pm-coach skill without an “unknown skill/agent” error.

Step 2 — Run `pm-setup`

First, cd into the folder you’ll use as your product workspace — an empty folder is fine if you have no existing code repo:

mkdir my-product-pm   # skip if you already have a project folder
cd my-product-pm
claude                # this opens a Claude Code session in the current directory

A “Claude Code session” is a CLI conversation started by running claude from inside a folder. The session knows which folder it’s in — that’s where pm-config.md will be written, and it becomes your workspace for the rest of the course.

In the session, invoke the setup skill:

Run the pm-setup skill to configure my notes system, issue tracker, and current desired outcome.

Answer its questions. Two notes:

For issue tracker, “none — I track work in plain notes/Markdown” is a valid answer. Don’t invent a tool you don’t use.
For current desired outcome, give the measurable behavior change you’re managing toward (Module 0 standard), not a feature. This is the single most important field — everything downstream inherits it.

Verify: a pm-config.md now exists in your workspace folder, and its fields match what you entered.

Step 3 — Wake up the PM coach

Start a fresh thread and ask a PM question:

Help me frame what I learned from a couple of customer conversations this week.

Verify: the pm-coach skill engages as a thinking partner — asking about outcomes and evidence — rather than just answering generically. (If it answers generically, the usual cause is a missing or empty pm-config.md — re-run Step 2.)

Step 4 — Confirm the core is real

Re-read your pm-config.md one more time and confirm the desired outcome field is a measurable behavior change, not a feature.

Verify: you could hand pm-config.md to a colleague and they’d know what behavior your product is trying to change.

✅ At this point the required core is done — you can complete the entire curriculum. Everything below is optional polish.

Recommended stack (optional)

Step 5 — Set up your notes + connections (as needed)

Full walkthrough: the PM Workspace Setup guide covers every step here in detail — one-command vault installer, exact BRAT install paths, how to add secrets, and JIRA/Linear API key setup. If you’re doing the full stack, follow that guide and come back here when it’s done.

Notes home — Obsidian + Claude Threads. Install Obsidian, then the Claude Threads plugin so Claude can work inside your notes. ⚠️ Confirm: the exact install channel for the Claude Threads and Vault Bridges plugins (Obsidian community store vs. manual/BRAT install) depends on how they’re distributed — check the plugin’s own README for the current method rather than assuming the community store.
Issue tracker (skip if you use plain notes). If you use Linear / Jira / JPD, connect it per the PM Tool Integration Guide. If you track work in Markdown, mark this N/A and move on — every exercise in this course works tool-agnostically.
Vault Bridges (optional). Bridge your product’s repo or a plain docs folder into your notes so product context is searchable. You don’t need a code repo — a folder of docs is fine.

Verify (only the pieces you set up): you can start a Claude thread inside Obsidian; if you connected a tracker, you can find one real issue from your notes; if you bridged a folder, you can find one real product file from your notes.

Deliverable: a working core environment and a friction log — a short list of every place a step failed, was unclear, named something wrong, or assumed context you didn’t have. (An empty friction log is a valid, great result — but only if you actually hit nothing.)

Success criteria

setup.sh completed and the agents + PM skills are available after a restart.
pm-config.md exists and its desired outcome is a measurable behavior change (Module-0 standard), not a feature.
pm-coach responds as a coach in a thread — it asks about outcomes/evidence.
You made a deliberate decision about each optional stack piece (set up or skipped on purpose) — a skipped tracker with “N/A” recorded counts as done.
You produced a friction log (even if it’s “no friction”).

Common failure modes

Symptom	What’s going wrong	Fix
`git clone` or `./setup.sh` errors	Tooling missing, or wrong directory	Confirm git and Claude Code are installed; run `setup.sh` from inside the cloned repo; use `--dry-run` to see what it expects. Log the exact error.
Skills/agents don’t appear (“unknown skill”)	Claude Code not restarted, or symlinks skipped	Restart Claude Code. Re-run `./setup.sh --force` if a previous install left stale entries. If it persists, log it with the exact command and output.
You tried the `claude plugins` command and the syntax was wrong	Plugin CLI differs by version; repo manifest is a subset	Use the `setup.sh` path — it’s the supported full install. Note the discrepancy in your friction log.
`pm-coach` answers generically	No `pm-config.md`, or it’s empty/stale	Re-run `pm-setup`; the coach anchors to your config.
`desired outcome` is a feature (“ship onboarding v2”)	Module-0 lesson didn’t carry over	Rewrite as a measurable behavior change (“raise day-2 activation from X% to Y%”). Every skill inherits this.
You feel blocked because you don’t have Linear / a code repo	Treating an optional piece as required	The tracker and vault bridge are optional. Record “N/A” and continue — the core is all the course needs.
You “just got it working” but skipped the friction log	Defeats the smoke-test purpose	Reconstruct it now. The next adopter and the public install path depend on what you hit.

Core’s live. Time to build the structure the whole operating model hangs on — your first Opportunity Solution Tree.

→ Module 2: Your First OST