Complete guide for Claude Code (CLI), Claude Desktop, and Claude Web.
| Tool | Best For | Setup |
|---|---|---|
| Claude Code (CLI) | Power users, best quality, multi-skill workflows | Medium |
| Claude Desktop | Everyday PM work, file uploads, reusable projects | Low |
| Claude Web | Quick one-off tasks, no installation needed | None |
Claude Code gives you the best output quality because it can read skill files directly from your local filesystem.
# 1. Clone the repo
git clone https://github.com/nandanosql/pm-skills.git
cd pm-skills
# 2. Start Claude Code in the repo directory
claude
# 3. Ask Claude to use a skill
> Use skills/discovery/problem-statement/SKILL.md to help me frame
our checkout abandonment problem.Single skill:
Read skills/planning/prioritization-advisor/SKILL.md and use it to help me
prioritize our Q3 backlog.
Multiple skills in sequence:
First, read skills/discovery/problem-statement/SKILL.md and help me define the problem.
Then use skills/specification/prd-development/SKILL.md to turn it into a PRD.
Workflow skill (orchestrates others automatically):
Run the discovery process using skills/discovery/discovery-process/SKILL.md.
It will pull in related skills as needed.
- Point Claude at the repo root — it can navigate to any skill file
- Use trigger phrases — each skill has triggers in its YAML frontmatter that activate it naturally
- Ask for quality scoring — say "score this output against the quality rubric" after generation
- Chain skills explicitly — "Now use [next skill] to continue" for multi-step workflows
- Open a new Claude conversation
- Click the 📎 attachment icon
- Upload one or more
SKILL.mdfiles from the repo - Tell Claude to follow the skill:
I've uploaded a PM skill file. Use it as the operating standard for our session.
Follow each section in order: Purpose → Key Concepts → Application → Examples.
Ask me up to 3 clarifying questions before drafting.
- Open the
SKILL.mdfile in GitHub or your editor - Copy the entire contents
- Paste into the Claude chat window
- Add your request after the pasted content
- Create a new Claude Project
- Add skill files to the project's Knowledge Base
- Set the project instructions:
When performing PM tasks, use the skill files in this project's knowledge base
as operating standards. Follow each skill's sections in order: Purpose, Key Concepts,
Application, Examples, Common Pitfalls, References. If context is missing, ask up
to 3 clarifying questions before drafting. Self-score outputs against the Quality
Rubric and flag any dimension scoring below 2.
- Every conversation in this project will automatically use your loaded skills
Use this in any Claude setup for consistent results:
## PM Skills Operating Instructions
1. When I reference a PM skill, follow its sections in strict order:
Purpose → Key Concepts → Application → Examples → Quality Rubric → Pitfalls
2. For Interactive skills: ask all clarifying questions BEFORE generating output.
3. For Component skills: generate the complete template, then self-score.
4. For Workflow skills: announce each phase and which sub-skills you're using.
5. After generating any artifact, score it against the Quality Rubric (target: 8+/10).
If any dimension is below 2, suggest specific improvements.
6. Flag Common Pitfalls proactively — don't wait for me to ask.
| Problem | Fix |
|---|---|
| Claude ignores the skill structure | Re-paste the instruction snippet above at the start of the conversation |
| Output is too generic | Add project-specific context: your product, market, user segments |
| Claude skips the quality rubric | Explicitly ask: "Score this against the quality rubric" |
| File upload not working | Try copy-paste instead — some Claude plans have upload limits |
| Too many skills loaded at once | Start with 1-3 focused skills; more can dilute quality |
You: [Uploads prioritization-advisor SKILL.md]
Help me prioritize features for our B2B SaaS product's Q3 roadmap.
Claude: I'll use the Prioritization Advisor skill to help you.
First, let me understand your context:
1. What type of decisions are you prioritizing?
(New features / Tech debt / Bug fixes / Mixed backlog)
2. How many items need prioritization? (~10 / ~25 / 50+)
3. Who will consume this prioritization?
(Engineering team / Leadership / Board)
4. Do you have data on: revenue impact, reach, effort estimates?
Based on your answers, I'll recommend RICE, ICE, Kano, MoSCoW,
or WSJF and walk you through applying it.
You: Mixed backlog of about 25 items, for leadership.
We have rough effort estimates but limited revenue data.
Claude: Given a mixed backlog with limited revenue data, I recommend
the ICE framework (Impact, Confidence, Ease)...
[Continues with structured guidance]