// the kit · what you cloned

What you cloned.

$tree .claude/skills/ references/ knowledge/
// 01 · what it is

A system, not a chatbot.

The PM Agent Kit encodes the accumulated judgment of a senior product manager as a set of portable, invocable skills that run in Claude Code. It’s not a collection of prompts or templates — it’s a system: each skill reads your company context, applies PM judgment from a shared reference library, and produces structured, reviewable artifacts.

The kit is designed to travel. Everything in .claude/skills/ and references/ is company-agnostic — it encodes how a strong PM thinks, not what any specific company does. Company-specific context lives in company/ and gets rebuilt at each new role. Your accumulated work lives in knowledge/ and follows you too.

The workbook is the structured practice companion. Each day introduces one skill against a consistent fictional product — Terrain — so you can see exactly how the kit performs at full capacity before applying it to your real work.

// 02 · the skills

Eighteen invocable skills. One for every PM job.

Week 01 Writing & thinking
01Writing
/prd-draft
Turn any rough input into a complete, structured PRD
02Writing
/doc-review
Evaluate a PM document against expert quality criteria
03Writing
/generate-tasks
Break a spec into user stories with acceptance criteria
04Writing
/sprint-plan
Assess backlog health and optionally draft a sprint plan
05Writing
/status-update
Synthesize delivery health into a stakeholder-ready update
Week 02 Evidence & discovery
06Evidence
/discovery-plan
Map assumptions by risk and plan the cheapest credible validation
07Evidence
/user-feedback
Cluster customer feedback into themes, severity, and implications
08Evidence
/data-analysis
Interpret product metrics in context and generate visualizations
09Evidence
/competitive-intel
Read the competitive landscape and translate signals into implications
10Evidence
/business-case
Build the investment argument: problem, impact, cost, risks
Week 03 Strategy & alignment
11Strategy
/roadmap-prioritization
Sequence initiatives and produce a defensible rationale
12Strategy
/decision-log
Capture decisions with context and reasoning that stays findable
13Strategy
/alignment-memo
Codify team operating norms and escalation paths
14Strategy
/meeting-brief
Prepare for a specific meeting with context, agenda, and decisions
15Strategy
/retro-synthesis
Find patterns in retrospective feedback before they become permanent
Week 04 Comms & integration
16Comms
/launch-checklist
Generate a complete checklist covering all launch readiness dimensions
17Comms
/one-pager
Compress a complex initiative into a single-page executive summary
18Comms
/presentation-deck
Build a narrative deck and export to HTML, PDF, or PPTX
// 03 · reference library

26 files of encoded PM judgment.

These aren’t documentation — they’re the judgment that runs underneath every skill. pm-philosophy.md is a good example: it encodes the reasoning behind every quality criterion in the system, and skills across all four weeks load it to understand why a standard exists, not just what it is. The same file shapes how a PRD gets drafted, how a retro gets synthesized, and how a business case gets structured.
FileWhat it encodes
PM Judgment
pm-philosophy.mdThe reasoning behind every quality criterion in the system
pm-smell-test.mdRed flags that signal an artifact isn’t ready to move forward
decision-frameworks.mdHow to structure decisions that stay legible at six months
prioritization-judgment.mdHow to sequence competing initiatives with transparent reasoning
pushback-and-negotiation.mdHow to protect scope, navigate disagreement, and escalate when needed
Quality Criteria
quality-criteria-prd.mdWhat a strong PRD must satisfy across problem, solution, and data dimensions
quality-criteria-tech-spec.mdWhat a good engineering spec must surface for PM review
quality-criteria-ticket.mdWhat a good user story or ticket must contain
quality-criteria-project-brief.mdQuality bar for project briefs calibrated to maturity stage
quality-criteria-general-document.mdSix universal quality dimensions for any PM document
acceptance-criteria.mdAC standards written for AI coding agent implementers
Communication & Structure
communication-quality.mdWhat makes PM communication worth reading across all formats
narrative-structure.mdHow to build a presentation arc where each slide earns the next
story-structure.mdHow to scope and organize stories for AI coding agent implementers
audience-registers.mdBehavioral profiles for adjusting depth and tone by audience type
slide-design.mdHow to compose slides for hierarchy, density, and visual clarity
visualization-standards.mdHow to select chart types and render findings that carry the insight themselves
branding-guidelines.mdWhat brand consistency looks like in presentations and how to apply it
Research & Analysis
discovery-methods.mdHow to identify assumptions, rank by risk, and plan credible validation
user-feedback-analysis.mdHow to synthesize feedback into what customers need, not just what they said
data-interpretation.mdHow to turn metrics into product decisions, not reports
competitive-analysis.mdHow to translate market signals into product implications
business-case-standards.mdHow to argue whether to invest in an initiative and at what level
launch-readiness.mdWhat a complete launch covers across all readiness dimensions
sprint-planning.mdWhat makes a sprint plan and backlog ready for the team
Output Format
agent-readable-output.mdShared vocabulary for structured output blocks across all skills
// 04 · knowledge

Artifacts accumulate. The kit gets richer over time.

Every skill that produces a document saves its output to a typed folder inside knowledge/. The folder isn’t just storage — it’s context. When /doc-review evaluates a PRD, it can read prior PRDs in knowledge/prds/ to understand your norms. When /sprint-plan drafts a new plan, it can read prior plans to calibrate velocity assumptions. The kit learns from its own output.

DirectoryWritten by
knowledge/prds//prd-draft
knowledge/tasks//generate-tasks
knowledge/sprint-plans//sprint-plan (Draft mode)
knowledge/status-updates//status-update (Draft mode)
knowledge/decisions//decision-log
knowledge/meeting-briefs//meeting-brief
knowledge/memos//alignment-memo
knowledge/discovery-plans//discovery-plan
knowledge/user-feedback//user-feedback
knowledge/data-analyses//data-analysis
knowledge/competition//competitive-intel
knowledge/business-cases//business-case
knowledge/roadmaps//roadmap-prioritization
knowledge/retros//retro-synthesis
knowledge/launch-checklists//launch-checklist
knowledge/one-pagers//one-pager
knowledge/presentations//presentation-deck
// 05 · company context

Portable by design.

The company/ directory holds everything that changes when you change jobs: product facts, team structure, sprint process, tool configurations. Skills read these files to produce output that’s specific to your context rather than generic.

company/facts/Product areas, customers, team structure, glossary
company/norms/Sprint process, decision-making, communication patterns
company/interfaces/Tool configs — Jira, Slack, data sources, branding

When context files are missing or still contain placeholder text, each skill applies a declared degradation rule — either proceeding with a visible caveat or stopping to tell you what it needs. Output quality scales directly with context depth. The workbook’s Terrain files give every skill real, substantive context from Day 0.

// 06 · how skills work

Every skill follows the same execution model.

step.01→load context

Read company context and references

The skill reads any files it declared as context-required or context-optional. If a required file is missing or still contains placeholder text, the skill applies its degradation rule — proceeding with a caveat or stopping to explain what’s needed. It then loads the relevant files from references/ as its judgment criteria.

step.02→run intake

Assess input and adapt

Before executing, the skill assesses how much signal your input provides. Rich input (problem, audience, and constraints are clear): restates understanding and proceeds. Moderate input (some gaps): asks up to three targeted questions with specific options. Thin input (a sentence or a vague ask): presents a structured interpretation of what it’s inferring and asks you to confirm before continuing. The cap is four questions.

step.03→execute

Apply reference judgment and produce the artifact

The skill executes against your input using the loaded reference files as its standard. A /prd-draft generates toward quality-criteria-prd.md. A /doc-review evaluates against the same file. The same judgment that catches problems in a review is what prevents them during generation.

step.04→draft-confirm

You review before anything moves

All skills start at draft-confirm: the agent produces output, you review it before using it. Nothing is sent, shared, or published automatically. Autonomy is earned through consistent output quality over time — it’s graduated, not assumed.

← Setup Company Context →