Using Skills

This guide takes you from basic skill invocation to advanced workflows and power-user techniques.

---

Table of Contents

Beginner

• Understanding Skill Structure
• Basic Invocation
• Providing Good Context
• Working with Outputs

Intermediate

• Iterating on Outputs
• Chaining Skills
• Using Workflows
• Customizing Outputs

Advanced

• Building Custom Workflows
• Cross-Referencing Artifacts
• Integrating with Your Process
• Power User Patterns

Reference

• Skill Quick Reference
• Common Patterns by Role

---

Beginner

Understanding Skill Structure

Every skill has three components that work together:

1. SKILL.md . The Instructions

This file tells the AI:

• What to create (the artifact type)
• When to use it (context and triggers)
• How to create it (step-by-step process)
• Quality criteria (what good looks like)

Example from the prd skill:

## Instructions

When asked to create a PRD, follow these steps:

1. **Summarize the Problem**
   Start with a brief recap of the problem being solved...

2. **Define Goals and Success Metrics**
   Articulate what success looks like...

2. TEMPLATE.md . The Structure

This file provides the exact format for the output:

• Section headings
• Required fields
• Formatting conventions

The AI uses this to ensure consistent, complete outputs.

3. EXAMPLE.md . The Quality Benchmark

A real-world example showing what good looks like:

• Proper tone and detail level
• How to handle ambiguity
• What completeness looks like

Key insight: You don’t need to read these files.the AI does. But understanding they exist helps you know what to expect.

---

Basic Invocation

The Simplest Way

Just ask for what you need:

Create a PRD for a search feature

The AI will find and use the prd skill automatically (if skills are loaded).

Explicit Skill Reference

For more control, reference the skill directly:

Use the prd skill to create a Product Requirements Document
for adding search to our e-commerce platform

Slash Commands (Claude Code)

The fastest method if you’re using Claude Code:

/prd search feature for e-commerce platform

What Happens Behind the Scenes

Your Request
     │
     ▼
┌─────────────────────────────────────┐
│ AI reads SKILL.md                   │
│ - Understands what to create        │
│ - Follows step-by-step instructions │
└─────────────────────────────────────┘
     │
     ▼
┌─────────────────────────────────────┐
│ AI reads TEMPLATE.md                │
│ - Applies correct structure         │
│ - Includes all required sections    │
└─────────────────────────────────────┘
     │
     ▼
┌─────────────────────────────────────┐
│ AI references EXAMPLE.md            │
│ - Calibrates quality level          │
│ - Matches tone and depth            │
└─────────────────────────────────────┘
     │
     ▼
Your Output (structured PM artifact)

---

Providing Good Context

The quality of your output depends heavily on the context you provide. Here’s how to give great context:

The SPICE Framework

Use this checklist to provide complete context:

Letter	Element	Example
S	Situation	”We’re an e-commerce platform with 50k daily users”
P	Problem	”Users can’t find products.search is broken”
I	Impact	”We’re losing $50k/month in missed sales”
C	Constraints	”Must ship in 6 weeks, can’t change database”
E	Evidence	”User interviews show 70% frustration rate”

Bad vs. Good Context

Bad (vague):

/prd search feature

Better (some context):

/prd search feature for our e-commerce site

Good (SPICE context):

/prd

We need a search feature for our e-commerce platform (50k daily users).

Situation: Currently users can only browse by category
Problem: Users report they can't find specific products
Impact: Support tickets mention "can't find" 200+ times/month
Constraints: Must work with existing Elasticsearch cluster, ship in Q2
Evidence: User interviews show 70% frustration when looking for specific items

Context Prompts

If you don’t provide enough context, expect the AI to ask questions like:

• Who are the target users?
• What’s the current state?
• What constraints exist?
• How will success be measured?

Pro tip: Pre-answer these questions in your initial request to get a complete output faster.

---

Working with Outputs

Understanding Output Structure

Every skill produces a structured artifact. Here’s what to expect:

Skill	Primary Output	Key Sections
problem-statement	Problem framing doc	User segment, Pain points, Success metrics
prd	Requirements doc	Problem, Solution, Requirements, Scope
user-stories	Story set	Stories, Acceptance criteria, Dependencies
hypothesis	Hypothesis statement	Belief, Metric, Threshold
experiment-design	Test plan	Hypothesis, Variants, Sample size, Duration

Reading Skill Outputs

Outputs follow a consistent pattern:

# [Artifact Title]

## Overview / Summary
High-level context for quick scanning

## Core Content
The main substance (requirements, insights, etc.)

## Supporting Details
Dependencies, risks, considerations

## Next Steps / Actions
What to do with this artifact

Validating Output Quality

Each skill includes a quality checklist. Use it:

## Quality Checklist (from prd skill)

- [ ] Problem and "why now" are clearly articulated
- [ ] Success metrics are specific and measurable
- [ ] Scope boundaries are explicit (in/out/future)
- [ ] Requirements are testable and unambiguous
- [ ] Technical considerations are surfaced
- [ ] Dependencies and risks are documented

Action: After receiving output, mentally check these boxes. Ask for revisions if anything is missing.

---

Intermediate

Iterating on Outputs

Your first output is rarely perfect. Here’s how to refine it:

Targeted Feedback

Be specific about what to change:

The success metrics section is too vague.
Make them more specific with baselines and targets.

Instead of:

Make it better

Section-by-Section Refinement

Focus on one section at a time:

The scope section looks good.
Now expand the technical considerations.we have
specific constraints around our API rate limits.

Adding Missing Context

If the output misses something, add the context:

Good start. I should mention we have an existing
recommendation engine. Update the PRD to integrate
with it rather than building new ML models.

Tone and Audience Adjustment

This PRD is too technical for the executive audience.
Rewrite the overview section for non-technical stakeholders,
keeping the detailed sections as-is for engineering.

Common Iteration Patterns

Issue	Iteration Prompt
Too vague	”Make [section] more specific with concrete examples”
Too detailed	”Summarize [section] to key points only”
Wrong audience	”Rewrite for [audience] at [detail level]“
Missing section	”Add a section covering [topic]“
Wrong tone	”Make this more [formal/casual/technical/accessible]“

---

Chaining Skills

Skills are designed to work together. The output of one becomes input to the next.

Natural Skill Chains

problem-statement → hypothesis → prd → user-stories
     (why)           (belief)   (what)    (how)

How to Chain

Method 1: Reference Previous Output

Step 1:
/problem-statement "checkout abandonment issue"

[AI produces problem statement]

Step 2:
Based on the problem statement above, use the hypothesis skill
to create a testable hypothesis.

[AI uses context from previous output]

Method 2: Explicit Handoff

Step 1:
/problem-statement "checkout abandonment issue"

[Save or note the output]

Step 2:
/hypothesis

Use this problem statement as context:
[Paste key points from previous output]

Method 3: Single Conversation Chain

Keep everything in one conversation thread:

You: /problem-statement "checkout abandonment"
AI: [produces problem statement]

You: Now create a hypothesis from this
AI: [produces hypothesis using conversation context]

You: Now write a PRD based on the hypothesis
AI: [produces PRD referencing both previous artifacts]

Recommended Chains

Goal	Chain
Feature development	problem-statement → hypothesis → prd → user-stories
Experiment planning	hypothesis → experiment-design → instrumentation-spec
Launch preparation	prd → edge-cases → launch-checklist → release-notes
Learning cycle	experiment-results → lessons-log → pivot-decision
Technical decisions	spike-summary → adr → design-rationale

Chain Dependencies

Some skills have natural dependencies:

                    ┌─────────────────┐
                    │ problem-statement│
                    └────────┬────────┘
                             │
              ┌──────────────┼──────────────┐
              ▼              ▼              ▼
        ┌──────────┐  ┌──────────┐  ┌──────────────┐
        │hypothesis│  │jtbd-canvas│  │opportunity-tree│
        └────┬─────┘  └──────────┘  └──────────────┘
             │
             ▼
        ┌─────────┐
        │   prd   │
        └────┬────┘
             │
    ┌────────┼────────┐
    ▼        ▼        ▼
┌────────┐ ┌────────┐ ┌────────────┐
│user-   │ │edge-   │ │instrumenta-│
│stories │ │cases   │ │tion-spec   │
└────────┘ └────────┘ └────────────┘

---

Using Workflows

Workflows are pre-defined skill sequences for common PM processes. There are 9 workflows covering feature development, research, strategy, experimentation, and more. Each can be invoked with a /workflow-{name} slash command (e.g., /workflow-feature-kickoff "launch mobile app").

For detailed guidance on choosing, invoking, customizing, and building workflows, see Using Workflows.

---

Customizing Outputs

Adjusting Detail Level

More detail:

/prd "search feature"

Make this PRD comprehensive.we need detailed requirements
for a large engineering team and external stakeholders.

Less detail:

/prd "search feature"

Keep this PRD lightweight.it's a small feature and
we need to move fast. Focus on essentials only.

Changing Output Format

For different audiences:

/prd "search feature"

Format the output for:
1. Executive summary (1 paragraph) at the top
2. Full details below for engineering

For different tools:

/user-stories "from the PRD above"

Format stories for Jira import with these fields:
- Summary (title)
- Description (full story)
- Acceptance Criteria (in Gherkin format)

Adding Custom Sections

/prd "search feature"

Add a section called "Competitive Context" that compares
our approach to how competitors handle search.

Using Your Templates

/prd "search feature"

Use this custom template structure instead of the default:

# [Feature Name] PRD
## TL;DR
## Problem
## Solution
## Metrics
## Timeline

---

Advanced

Building Custom Workflows

Creating Workflow Templates

Define your own skill sequences:

# My Discovery Workflow

1. stakeholder-summary (identify who matters)
2. interview-synthesis (understand users)
3. competitive-analysis (understand market)
4. problem-statement (frame the problem)
5. opportunity-tree (map solutions)

Pause for team review, then continue to:

6. hypothesis (what we'll test)
7. solution-brief (proposed approach)

Conditional Workflows

Build decision points into your workflow:

Start with problem-statement.

If technical uncertainty is high:
  → Run spike-summary before prd

If UX complexity is high:
  → Run design-rationale before prd

If experiment needed:
  → Add experiment-design after prd

Always end with:
  → launch-checklist before shipping

Parallel Workflows

Some skills can run in parallel:

After prd is complete, run in parallel:
- user-stories (for engineering)
- instrumentation-spec (for data team)
- edge-cases (for QA)

Then converge at launch-checklist.

Workflow Automation Prompt

I want to run a custom workflow. For each skill:
1. Generate the artifact
2. Show me a summary
3. Ask if I want to revise before continuing
4. Only proceed to next skill after I confirm

Skills in order:
1. problem-statement
2. hypothesis
3. solution-brief
4. prd

Context for all: [your feature context]

---

Cross-Referencing Artifacts

Building an Artifact Network

As you create artifacts, they should reference each other:

PRD:
  "See Problem Statement (link) for detailed user research"
  "Success metrics derived from Hypothesis (link)"

User Stories:
  "Implements PRD requirement R-001"
  "See Edge Cases (link) for error handling"

Launch Checklist:
  "Verify all PRD requirements are met"
  "Confirm instrumentation from Instrumentation Spec (link)"

Maintaining Consistency

When updating one artifact, update related ones:

I'm changing the success metric in the hypothesis from
"reduce abandonment by 15%" to "reduce abandonment by 20%".

Please update the following to match:
- The PRD success metrics section
- The experiment design primary metric
- The dashboard requirements KPI definition

Version Control for Artifacts

This is version 2 of the PRD.

Changes from v1:
- Added offline mode requirement (stakeholder feedback)
- Removed social sharing (descoped)
- Updated timeline to Q3

Please regenerate user stories reflecting these changes,
marking which stories are new, modified, or removed.

---

Integrating with Your Process

Mapping Skills to Your Workflow

Your Process Step	Relevant Skills
Discovery kick-off	stakeholder-summary, interview-synthesis
Problem review	problem-statement, opportunity-tree
Solution design	hypothesis, solution-brief, design-rationale
Spec review	prd, adr
Sprint planning	user-stories, edge-cases
Launch prep	launch-checklist, release-notes
Retro	retrospective, lessons-log

Integrating with Existing Templates

If you have existing templates, use them:

/prd "search feature"

We use a custom PRD template at our company.
Generate content following OUR template structure:

[Paste your company's PRD template]

Fill in each section following the prd skill's guidance
for what content belongs in each section.

Feeding Into Other Tools

For Jira/Linear:

/user-stories "from PRD above"

Format for bulk import to Jira:
- CSV format
- Columns: Summary, Description, Acceptance Criteria, Story Points, Labels
- Use label "search-feature" for all stories

For Notion/Confluence:

/prd "search feature"

Format with:
- Notion-compatible markdown
- Collapsible sections for details
- Table format for requirements
- Callout blocks for risks

For Slides:

/problem-statement "checkout abandonment"

Also create a 3-slide summary:
1. The Problem (with key stat)
2. The Impact (business case)
3. Success Criteria (metrics)

---

Power User Patterns

Pattern 1: The Pre-Flight Check

Before starting any skill, gather context:

Before I run /prd, help me prepare:

1. What information do you need from me?
2. What questions should I answer upfront?
3. What artifacts should exist before a PRD?

My context: [brief description]

Pattern 2: The Batch Generator

Generate multiple artifacts efficiently:

I'm kicking off 3 features this sprint. For each, generate:
1. A brief problem statement (2-3 sentences)
2. A hypothesis (1 sentence)
3. A scope summary (in/out bullets)

Features:
A. Dark mode support
B. Export to PDF
C. Keyboard shortcuts

Format as a comparison table so I can prioritize.

Pattern 3: The Quality Audit

Use skills to review existing work:

Here's our existing PRD for the search feature:
[paste PRD]

Using the prd skill's quality checklist, audit this PRD:
1. What sections are missing or incomplete?
2. What doesn't meet the quality bar?
3. What specific improvements would you make?

Pattern 4: The Reverse Engineer

Extract structure from examples:

Here's a PRD that our team loved:
[paste excellent PRD]

Analyze what makes it effective and create a custom
template that captures its best qualities, merged with
the standard prd skill template.

Pattern 5: The Stakeholder Variants

Generate audience-specific versions:

From this PRD, create three versions:

1. Executive Brief (1 page)
   - Problem/opportunity
   - Investment required
   - Expected return

2. Engineering Spec (detailed)
   - Technical requirements
   - API contracts
   - Performance criteria

3. Design Brief (for designers)
   - User journeys
   - Key screens
   - Interaction patterns

Pattern 6: The Living Document

Keep artifacts updated as you learn:

Here's our experiment-results document from last week's A/B test.
The results invalidated our hypothesis.

Please:
1. Update the hypothesis document to reflect learnings
2. Generate a pivot-decision document
3. If we pivot, draft a revised problem-statement
4. Outline what the new PRD scope would look like

Pattern 7: The Skill Combo

Combine skills for richer outputs:

Create a combined document that includes:

1. Problem Statement (brief, 1 section)
2. Hypothesis (1 section)
3. Solution Brief (1 section)
4. PRD (full)
5. User Stories (summary table)

All in one document with clear section breaks.
This is our "Feature One-Pager" format.

---

Reference

Skill Quick Reference

By Phase

Phase	Skills	Common Chain
Discover	interview-synthesis, competitive-analysis, stakeholder-summary	stakeholder → interview → competitive
Define	problem-statement, hypothesis, opportunity-tree, jtbd-canvas	problem → hypothesis
Develop	solution-brief, spike-summary, adr, design-rationale	spike → adr → solution
Deliver	prd, user-stories, acceptance-criteria, edge-cases, launch-checklist, release-notes	prd → stories → acceptance-criteria → edge-cases → launch
Measure	experiment-design, instrumentation-spec, dashboard-requirements, experiment-results	design → instrument → results
Iterate	retrospective, lessons-log, refinement-notes, pivot-decision	results → retro → lessons → pivot

By Output Type

Need	Skill	Output
Define the problem	problem-statement	Problem framing document
Test an assumption	hypothesis	Testable hypothesis
Pitch a solution	solution-brief	One-page solution overview
Specify requirements	prd	Product Requirements Document
Plan development	user-stories	User stories with acceptance criteria
Define done conditions	acceptance-criteria	Given/When/Then acceptance criteria
Handle edge cases	edge-cases	Edge case catalog
Plan an experiment	experiment-design	A/B test plan
Track events	instrumentation-spec	Analytics event specification
Prepare for launch	launch-checklist	Cross-functional checklist
Communicate changes	release-notes	User-facing release notes

---

Common Patterns by Role

Product Manager

Daily workflow:

Monday:    /refinement-notes (capture planning outcomes)
Tuesday:   /user-stories (prepare for sprint)
Wednesday: /prd (spec new features)
Thursday:  /experiment-results (review test outcomes)
Friday:    /retrospective (team reflection)

Key chains:

• Discovery: stakeholder-summary → interview-synthesis → problem-statement
• Delivery: prd → user-stories → launch-checklist

Engineering Lead

Key skills:

• adr . Document architecture decisions
• spike-summary . Capture exploration findings
• edge-cases . Ensure comprehensive coverage
• instrumentation-spec . Define tracking requirements

Common workflow:

/spike-summary "evaluate new database options"
/adr "based on spike findings, recommend PostgreSQL"
/edge-cases "for the migration plan"

Designer

Key skills:

• design-rationale . Document design decisions
• jtbd-canvas . Understand user motivations
• interview-synthesis . Process user research

Common workflow:

/interview-synthesis "from last week's user sessions"
/jtbd-canvas "for the checkout flow"
/design-rationale "for one-page checkout decision"

Data/Analytics

Key skills:

• experiment-design . Plan rigorous tests
• instrumentation-spec . Define event tracking
• dashboard-requirements . Specify reporting needs
• experiment-results . Document findings

Common workflow:

/experiment-design "test new pricing page"
/instrumentation-spec "for pricing experiment"
/dashboard-requirements "pricing experiment monitoring"
[after test]
/experiment-results "pricing experiment concluded"

---

Next Steps

Beginner → Intermediate

• Use 3 different skills successfully
• Chain 2 skills together in one conversation
• Iterate on an output to improve quality
• Try one workflow

Intermediate → Advanced

• Create a custom workflow for your team
• Generate artifacts for different audiences from one source
• Integrate skill outputs with your existing tools
• Use skills to audit/improve existing documents

Mastery

• Define team-specific skill chains
• Create custom templates that merge with skills
• Build artifact networks with cross-references
• Teach others to use skills effectively

---

See Also

• Getting Started . Installation and setup
• Categories Reference . Skill organization
• Workflows . Pre-built workflows

---

Part of PM-Skills . Open source Product Management skills for AI agents