
retention-diagnostic-advisor
Diagnostic advisor that identifies why customers churn or disengage, then matches each cause to a fix.
- GitHub stars
- 8.1K
What it does
- Covers retention diagnosis for customers who leave or go quiet, mapping each cause to a suitable intervention
- Designed for product managers working through churn and engagement problems in their own products
Find why customers leave or go quiet, then pick the fix that fits the cause.
Add marketplace
/plugin marketplace add deanpeters/Product-Manager-SkillsRun it inside Claude Code once, then install the plugin below.
Install
/plugin install retention-diagnostic-advisor@pm-skillsRun it inside Claude Code after adding the marketplace.
Files
Discovered from sourceRetention Diagnostic Advisor
Purpose
Find why customers leave or go quiet, and match the fix to the cause. Retention programs fail when they treat all churn as one problem: a discount saves the customer who found it too expensive and does nothing for the one who never got started.
The sentence to keep in your head: churn is the last event, not the cause. The cause happened earlier: the customer never got value, was never the right customer, got value and then lost it, or something around them changed. Each needs a different fix.
Input
Works best with: What's churning or going quiet (which customers, since when) and what you already know about why.
Also useful: Retention or renewal numbers by cohort and segment, exit or cancellation reasons, time from sign-up to churn, support history of customers who left, and how long a typical contract runs.
Anything supplied with the invocation itself (text after the skill name, a pasted context dump, or
an appended ARGUMENTS: line) counts as answers already given. Use it and skip whatever it covers;
don't re-ask.
Arriving empty-handed? That works too. The skill asks what's happening and what you've seen, and works out which cause is most likely from your answers. If the data can't tell the causes apart, it says so and names the cheapest way to find out.
Example invocations:
Churn rose from 6% to 9% last quarter. Most leavers had under 90 days with us.Service-contract renewals on the NFA-200 base fell from 62% to 48%. Contracts run 3 years.
Key Concepts
Four Causes of Leaving
| Cause | What it looks like | Typical timing |
|---|---|---|
| Never activated | Customer signed up or bought but never got first value (activation-design) | Early: first weeks or months; usage never started |
| Never a fit | Wrong customer: wrong size, use case, or expectation set in the sale | Early to mid; leaves after a trial or first renewal |
| Value faded | Got value, then stopped: needs changed, usage dropped, a champion left | Mid to late; usage falls before the exit |
| Something changed | Budget cut, merger, a competitor, a regulation, a product being retired | Any time; often not about your product |
Most real churn is a mix, but one usually dominates. Find it before you act.
Read the Timeline, Not the Rate
The average churn rate hides the pattern. Look at when in the customer's life they leave (by months since start) and what they did before (usage curve, support contacts, champion changes). Early-life churn points to activation and fit. Churn after a long healthy period points to faded value or outside change. A usage drop that precedes cancellation by weeks is your earliest warning.
Segment Before You Conclude
Cut by customer size, industry, source (direct, partner, self-serve), plan, and the version they
bought. Retention that is 95% in one segment and 60% in another is two problems, and the weak
segment might be customers you never should have sold to (sales-value-narrative
decides who you tell the story to).
Stated Reasons Are Weak Evidence
"Too expensive" is the most common exit reason and the least informative: it often means "didn't see the value." Treat exit reasons as a starting question ("What did you expect it to do for you?"), not a finding, and compare them with behavior (did they ever activate? did usage fall first?).
Leading Signals
Watch the signals that move before the exit: usage trend, breadth of use (how many people or
features), support sentiment, champion turnover, and missed milestones. For contract businesses
add: engagement with the service, unused entitlements, and who the renewal decision-maker is. A
simple health view built from these is better than waiting for a cancellation. See
launch-metrics for leading and lagging pairs.
Who to Save and Who to Let Go
Not every customer is worth saving. Ask: would they have been a good customer if the cause were fixed? Saving a bad-fit customer at a discount trains the market and fills support with the wrong requests. Be explicit about the customers you'll try to keep, the ones you'll let go, and why.
Right-Size the Response
A small contract base needs a conversation with each at-risk account. A large self-serve base needs an automated health signal and a few targeted plays. Don't build a heavy program to solve a cause the data hasn't confirmed. Recommend a level and let the user move it.
Facilitation Source of Truth
Use workshop-facilitation as the interaction protocol:
heads-up, one question per turn, numbered options, and best-guess mode on request.
Default entry mode: Guided · set in pm-skills.config.yaml (library-wide default or a per-skill override); the user's own choice in the session always wins. On the first turn, offer all three modes, then open in the default without waiting for a pick: Guided asks the first question, Context dump invites the paste, Best guess asks for a one-sentence minimum and drafts with every assumption labeled.
Anti-Patterns (what this is NOT)
- Not a retention program. It diagnoses; the fix depends on the cause.
- Not a churn forecast. It explains past behavior and points to early signals.
- Not a discount strategy. Price is rarely the cause and is a costly cure.
Application
This skill asks up to 4 adaptive questions, then recommends a diagnosis and a move.
Question 1 of 4: What's happening?
Agent asks:
"Describe what you're seeing: who is leaving or going quiet, since when, and what you've noticed about them."
Listen for the timeline and the segment. "Mostly under 90 days" and "mostly large accounts at renewal" point to different causes.
Question 2 of 4: When in their life do they leave?
Agent asks:
"When do customers typically leave?
- Early: before or just after they really start (first weeks or months)
- At the first renewal or trial end
- After a long, healthy period
- No clear pattern / I don't know"
Map to the likely cause: 1 → never activated. 2 → never a fit or never activated. 3 → value faded or something changed. 4 → say the data can't tell yet and name how to find out.
Question 3 of 4: Did they ever get started?
Agent asks:
"For the customers who left, what do you know about what they did before they left?
- Most never really used it (little or no usage)
- They used it, then usage dropped over weeks or months
- They used it until something changed (a person left, a merger, a budget cut)
- I haven't looked at this yet"
Confirm or revise. 1 → activation. 2 → value faded. 3 → something changed. 4 → recommend this check first; it's the cheapest way to separate the causes.
Question 4 of 4: Who are they?
Agent asks:
"Do the leavers share anything: size, industry, how they came to you, what they bought?"
Find the segment. A shared trait suggests a fit problem. No shared trait points to a product or service cause. Ask how big the base is, so the response can be sized.
Final Step: Recommend the diagnosis and move
## Retention Diagnosis: [context]
**Most likely cause:** [Never activated / Never a fit / Value faded / Something changed],
because [the evidence from their answers]
**Confidence:** [strong / suggestive / thin] · **What would change my mind:** [one check]
### The move
- **If never activated:** [fix the path to first value; see activation-design]
- **If never a fit:** [change who you sell to; adjust the story and qualification]
- **If value faded:** [reinforce value: usage reviews, new use cases, a champion plan]
- **If something changed:** [save plays only where it's winnable; plan for the rest]
### Leading signals to watch
[usage trend, breadth, support sentiment, champion turnover, unused entitlements]
### Who to save and who to let go
[with reasons]
### Next
1. [skill or action]
2. [skill or action]
Examples
Example A: SaaS, early churn
User: "Churn rose from 6% to 9% last quarter. Most leavers had under 90 days with us. Leadership wants a loyalty program."
Agent (Q2): → 1 (early). (Q3): → 1 (most never really used it). (Q4): "Mostly accounts under 100 technicians that came in through the March promo."
Diagnosis: Never activated, with a fit component. Confidence: suggestive (we have the
timing and the usage, not why). A loyalty program would target customers who never got started. The
move: fix the path to first value (activation-design) and look
at who the March promotion brought in. What would change my mind: if the leavers had real usage
before leaving.
Example B: Industrial, renewals
User: "Service-contract renewals on the NFA-200 base fell from 62% to 48%. Contracts run three years."
Agent (Q2): → 2 (at renewal). (Q3): → 3 (something changed): "Several plants were told the NFA-200 is being retired." (Q4): "Mostly plants with 1-2 units."
Diagnosis: Something changed: the product's end-of-sale announcement. Confidence: strong
(timing matches the announcement; the leavers share a trait). The move is not a loyalty program but
a clear transition path: contract terms through 2028, a migration offer for the NFA-500, and the
messaging from eol-message. Save plays for plants with 3+ units; let
the rest migrate on their schedule. What would change my mind: if non-announced plants also drop.
Common Pitfalls
Pitfall 1: Treating All Churn as One Problem
Symptom: One loyalty program for every leaver. Consequence: It misses the cause for most of them. Fix: Separate the four causes first.
Pitfall 2: Believing the Exit Survey
Symptom: "Too expensive" accepted as the cause. Consequence: Discounts that don't save the customer who never got value. Fix: Compare the stated reason with behavior.
Pitfall 3: Reading the Rate, Not the Timeline
Symptom: Tracking the average monthly churn. Consequence: Early-life and late-life problems blur together. Fix: Chart churn by months since start and by segment.
Pitfall 4: Saving Everyone
Symptom: Discounts to every at-risk account. Consequence: Bad-fit customers stay and drain support. Fix: Decide who is worth saving and why.
Pitfall 5: Waiting for the Cancellation
Symptom: Action begins at the cancel request. Consequence: The decision is already made. Fix: Watch leading signals: usage trend, breadth, champion changes.
Pitfall 6: Ignoring Outside Change
Symptom: Churn blamed on the product when a merger or announcement caused it. Consequence: A fix for something that wasn't broken. Fix: Check what changed around the customers.
References
Related Skills
activation-design: fix the path to first valueusage-data-investigation: read usage and cohortslaunch-metrics: leading and lagging indicators and triggerssaas-revenue-growth-metrics: churn and retention definitionseol-messageandeol-process: when retirement is the causeorganic-growth-advisor: growth paths beyond retentionworkshop-facilitation: interaction protocol
External Frameworks
- Cohort retention curves: standard product and customer analytics practice.
- Customer health scoring and leading signals: common customer-success practice.
- Value realization and time to value: customer-success literature.
Provenance
Distilled from practitioner experience diagnosing churn and renewal problems in software and industrial service businesses.
From the deanpeters/Product-Manager-Skills README
Product Manager Skills
Ready-made PM skills for your AI assistant. A skill is a short instruction file that teaches an AI how to do one PM job the way a pro would, such as writing a PRD or mapping stakeholders, so you stop re-explaining your process every time.
Try it in five minutes
- Download the starter pack and unzip it.
- In Claude, open Settings → Capabilities → Skills and upload one skill ZIP from inside the folder. Using Claude Code, Codex or something else? Jump to Get Started.
- Start a chat and paste this, swapping in your own details:
Which of these skills fits what's on my desk this week? Here's the situation: [one or two sentences].
You should see Claude name the skill it is using and start asking you questions. Arriving empty-handed is fine: the skills ask for what they need.
╔════════════════════════════════════════════════════════════════════╗
║ ║
║ ██████╗ ███╗ ███╗ ███████╗██╗ ██╗██╗██╗ ██╗ ███████╗
║ ██╔══██╗████╗ ████║ ██╔════╝██║ ██╔╝██║██║ ██║ ██╔════╝
║ ██████╔╝██╔████╔██║ ███████╗█████╔╝ ██║██║ ██║ ███████╗
║ ██╔═══╝ ██║╚██╔╝██║ ╚════██║██╔═██╗ ██║██║ ██║ ╚════██║
║ ██║ ██║ ╚═╝ ██║ ███████║██║ ██╗██║███████╗███████╗███████║
║ ╚═╝ ╚═╝ ╚═╝ ╚══════╝╚═╝ ╚═╝╚═╝╚══════╝╚══════╝╚══════╝
║ ║
║ 95 battle-tested skills + 6 command workflows ║
║ Claude Code • Cursor • Codex • n8n • OpenClaw • and more ... ║
║ ║
║ v0.86 • November 13, 2026 • CC BY-NC-SA 4.0 ║
╚════════════════════════════════════════════════════════════════════╝
95 battle-tested PM frameworks, ready for Claude, Codex, ChatGPT, and any agent that can read structured knowledge.
Why This Exists
Generic AI output is a PM's worst enemy. When you tell your agent "write a PRD" without shared context, you get a generic document that no stakeholder trusts and no engineer can act on.
This library gives both you and your AI agent the same professional foundation: the why behind each framework, the failure modes to avoid, and the judgment to apply them correctly. You stop repeating yourself. Your agent stops guessing. The work gets better.
The goal is dual — functional and pedagogic in equal measure. Skills equip agents to do PM work at a professional level, and they teach the human PM the reasoning behind each framework — so you can explain it, adapt it, and pass it on. Neither is a byproduct of the other.
Get Started
Choose your setup:
| I use... | Get this | Notes |
|---|---|---|
| Claude Desktop or Claude Web | pm-skills-starter-pack.zip | Unzip, then upload the individual skill ZIPs to Claude Skills |
| Claude Code | Plugin marketplace | /plugin marketplace add deanpeters/Product-Manager-Skills, then /plugin install product-manager-skills@pm-skills for everything, or one lifecycle bundle at a time (see Browse by lifecycle) |
| Codex | pm-skills-codex.zip | Installs .agents/skills and AGENTS.md |
| OpenCode | Copy skills/* to ~/.config/opencode/skills/ (or reuse the Codex ZIP) | OpenCode reads our SKILL.md format as-is. See the install guide |
| Not sure | pm-skills-starter-pack.zip | Start here |
All downloads: GitHub Releases
Start here by situation
Don't know where to begin? Find the line that sounds like your week. Each route is a suggested order, not a pipeline — every skill runs on its own, so skip ahead, reorder, or stop after one. The first prompt is real: paste it, swap in your details, and go.
| If this is you... | Try, in this order | Time | Paste this first |
|---|---|---|---|
| "I have an idea and I don't know if it's real." | problem-framing-canvas → discovery-interview-prep → pol-probe-advisor | ~1–1.5 hr | Frame this problem: field sales reps 'don't use the CRM' and leadership wants gamification. |
| "A request or mandate just landed on my desk." | incoming-request-advisor → stakeholder-mapping → prioritization-advisor | ~1 hr | My VP DM'd me: "Any chance the dashboard redesign lands next sprint? Board's asking." Analyze this before I reply. |
| "I need to understand the market, fast." | market-landscape-scan → competitive-research-snapshot → swot-analysis | ~1–2 hr | Run a market landscape scan on [your market], and tell me the decision it should support. |
| "We keep getting compared and losing." | positioning-workshop → roadmap-planning (going bigger: product-strategy-session, a 2–4 week workflow) | ~2–3 hr | Run a positioning workshop for our API monitoring tool — we keep getting compared to Datadog and losing. |
| "Time to write it down." | prd-development → epic-breakdown-advisor → user-story | ~1.5–3 hr | Build a PRD for self-serve workspace provisioning — here are my discovery notes and the OKR it ladders to. |
| "Our product is getting old." | lifecycle-play-advisor → eol-readiness-advisor → eol-process | ~30–50 min to a decision | Help me decide what to do with [product]. Here's what makes me think it's fading: [signals]. |
| "I'm stepping up a level." | director-readiness-advisor or vp-cpo-readiness-advisor — pick by where you're headed | ~10–20 min | I'm 60 days into a Director role and my old peers now report to me — it's awkward and I'm still doing PM work. Coach me. |
| "I'm new here, or not sure." | The Starter pack — six skills across the craft | — | Install it, then ask: Which of these skills fits what's on my desk this week? |
Why these orders (so you can break them on purpose):
- Idea → problem → people → probe. Name the problem before you ask anyone about it, and ask people before you build anything. The probe comes last because it's the most expensive of the three.
- Request → decode → map → decide. Understand what's really being asked before replying, know who else is in the room, then ask where it fits. Prioritizing first is how a loud VP becomes your roadmap.
- Market → competitors → SWOT. Evidence first, framework last. SWOT on top of no research is opinion with a 2×2 around it.
- Positioning → roadmap. Decide who you're for and why you win before choosing what to build. The strategy session is the full-length version when the question is bigger than a quarter.
- PRD → epics → stories. The why first, then slices, then the smallest writeable pieces. Writing stories first is how a backlog loses its reason.
- Aging → play → readiness → process. The first skill may say "nothing yet" — extending or harvesting is a legitimate answer, and starting here does not commit you to a sunset.
Browse by lifecycle
Product work unfolds in an order: set the outcome before you investigate, investigate before you define, define before you build. The library follows that arc in five bundles, with a sixth that runs underneath all of them. It's a cycle, not a pipeline, so you will loop back, and every skill still runs on its own.
| Bundle | The question it answers | Start with | Claude Code install |
|---|---|---|---|
| Determine | What are we trying to achieve, and what is the market doing? | product-strategy-session, market-landscape-scan, business-health-diagnostic | /plugin install pm-determine@pm-skills |
| Discover | What is the real problem? | problem-framing-canvas, discovery-interview-prep | /plugin install pm-discover@pm-skills |
| Define | Is this solution worth building, and how do we position it? | pol-probe-advisor, positioning-statement | /plugin install pm-define@pm-skills |
| Develop | How do we turn the chosen solution into buildable work? | prd-development, user-story, prioritization-advisor | /plugin install pm-develop@pm-skills |
| Deliver | How do we price it, take it to market, and retire it well? | finance-based-pricing-advisor, lifecycle-play-advisor | /plugin install pm-deliver@pm-skills |
| Drive | How do I work with stakeholders and grow my career? | stakeholder-mapping, director-readiness-advisor | /plugin install pm-drive@pm-skills |
Not sure which bundle? Ask the pm-skills-navigator: describe what's on your desk, and in about five minutes it tells you where the work really is and which skill to open first.
Every bundle includes a few shared helper skills (the facilitation rulebook and the navigator among them), so each works installed alone. The full list, grouped by bundle and by the smaller "motions" inside each, is in catalog/skills-by-bundle.md.
Why this order? Skipping ahead is how teams build the right thing for the wrong problem. Naming the outcome first gives discovery a target, and discovery gives definition something real to test.
On Claude Desktop or Web? Download the bundle pack that matches your phase, or grab single skills from the download shelf.
Lifecycle bundle packs for Claude Desktop / Web
Each pack is a ZIP of upload-ready skill ZIPs. Unzip it, then upload the skills to Claude Skills:
| Pack | Download | What's inside |
|---|---|---|
| Starter | pm-skills-starter-pack.zip | Six core skills across the craft. The best first download |
| Determine | pm-determine-pack.zip | Strategy, finance, market intelligence, customer context (24 skills) |
| Discover | pm-discover-pack.zip | Interviews, surveys, usage data, synthesis, assumptions, problem framing, opportunity trees, journey maps (12 skills) |
| Define | pm-define-pack.zip | Hypotheses, probes, experiments, positioning (9 skills) |
| Develop | pm-develop-pack.zip | PRDs, stories, prioritization, roadmaps, sprint and release rituals, AI agents (16 skills) |
| Deliver | pm-deliver-pack.zip | Launch plans, messaging, sales enablement, pricing, channels, post-launch metrics, activation and retention, and the end-of-life suite (18 skills) |
| Drive | pm-drive-pack.zip | Stakeholders, decision rights, influence, and career transitions (11 skills) |
| All skills | 99-all-skills-pack.zip | All 95 skills |
The older themed packs (02-discovery through 07-lifecycle-eol) were retired in v0.86. Every skill in them lives in a lifecycle pack above.
Install guides
Try It First — Streamlit (beta)
Not ready to wire skills into your agent setup? Run the local playground first and kick the tires in your browser.
pip install -r app/requirements.txt
streamlit run app/main.py
What you can do:
- Learn — browse setup and integration paths without leaving the app
- Find My Skill — describe your situation in plain English and get recommended skills
- Run Skills — run a skill with your own scenario once you know what you want
Multi-provider support: Anthropic, OpenAI, Ollama. API keys via environment variables only (no in-app key entry).
Docs: app/STREAMLIT_INTERFACE.md · app/.env.example
Feedback welcome via GitHub Issues or LinkedIn.
What You Can Get Done
Navigate by what you're actually trying to accomplish:
Framing and strategy
- problem-framing-canvas — MITRE's Look Inward / Look Outward / Reframe sequence; stops teams from solving the wrong problem
- positioning-statement — Geoffrey Moore's template for defining who you serve, what you solve, and how you're different
- product-strategy-session — full strategy arc: positioning → problem framing → solution exploration → roadmap (2-4 weeks)
Stakeholder alignment
- stakeholder-identification — map every stakeholder before engaging anyone: broad brainstorm → allies/audiences/influencers → R/P/D marking → equity lens → narrow to priority targets
- stakeholder-mapping — run two complementary grids (Power × Interest for engagement strategy; Impact × Power for whose voice to elevate) and compare to find the gaps
- stakeholder-engagement-advisor — per-stakeholder engagement planning: diagnoses their profile and context, then delivers tailored message framing, medium, cadence, and a named next action
Customer discovery and research
- discovery-interview-prep — plans Mom Test-style interviews based on your research goals
- opportunity-solution-tree — generates opportunities and solutions, then recommends the best proof-of-concept to test first
- discovery-process — complete discovery cycle: frame → research → synthesize → validate (3-4 weeks)
Prioritization and roadmapping
- prioritization-advisor — asks 3-5 questions about your context, then recommends RICE, ICE, Kano, or the right alternative
- epic-breakdown-advisor — splits large epics using Richard Lawrence's 9 patterns
- roadmap-planning — gather inputs → define epics → prioritize → sequence → communicate (1-2 weeks)
Writing PM deliverables
- user-story — Mike Cohn format + Gherkin acceptance criteria, with anti-patterns
- prd-development — structured PRD: problem → personas → solution → metrics → stories (2-4 days)
- press-release — Amazon Working Backwards: clarify product vision before writing a line of spec
Validation and experimentation
- pol-probe-advisor — recommends which prototype type to run based on your hypothesis and risk level
- pol-probe — template for documenting lightweight validation experiments before building
Finance and growth
- business-health-diagnostic — diagnoses SaaS health across growth, retention, efficiency, and capital using your real metrics
- organic-growth-advisor — McKinsey Growth Pyramid triage: diagnoses whether your constraint is in new segments, geographies, channels, or products
- feature-investment-advisor — build / don't build recommendation using revenue impact, cost, ROI, and strategic value
Market and competitive intelligence
- intel-discipline-advisor — start here: triages the question on your desk into the right intelligence disciplines, cadence, and executing skill
- competitive-analysis-process — the six-step umbrella: landscape → product comparison → customer needs → business baseline → positioning → strategic direction
- The investigation chain: market-landscape-scan → competitive-research-snapshot → competitive-intel-watch → battle-card-builder — each consumes the prior's schema, so research becomes a cadence instead of a one-off deck
- tam-sam-som-calculator — market sizing three ways: your own numbers, guided interview, or autonomous bottom-up research a skeptical CFO can attack one assumption at a time
Aging products: extend, replace, or retire
- lifecycle-play-advisor — start here when a product stops growing: seven transition questions establish the stage, then it picks the play — extend, replace, or retire — and tells you when the honest answer is "nothing yet"
- product-lifecycle-plays — the framework behind that call: the PLC strategy grid, the seven replacement hazards, and a risk register whose last column asks what is Plan B?
- eol-process — the whole sunset in six phases, decide → align → plan → prepare → announce → close, including the post-EOL review everyone skips
- The EOL chain: eol-readiness-advisor → eol-stakeholder-sequence → eol-checklist → eol-internal-enablement → eol-message — every one runs standalone, sized Light/Standard/Heavy by you
Career and leadership transitions
- director-readiness-advisor — coaches PMs through the PM→Director shift across four situations: preparing, interviewing, newly landed, recalibrating
- vp-cpo-readiness-advisor — coaches Directors through VP/CPO transition, including a CEO interview framework for evaluating roles before you accept
- executive-onboarding-playbook — 30-60-90 day diagnostic playbook for VP/CPO transitions
AI product work
- ai-shaped-readiness-advisor — assesses whether you're automating tasks (AI-first) or redesigning how you work (AI-shaped), then tells you which competency to build first
- context-engineering-advisor — diagnoses context stuffing vs. context engineering and guides memory architecture for AI-powered products
- agent-orchestration-advisor — guides multi-agent workflow design across four dimensions
95 Skills, 3 Types
Skills are organized in three tiers that build on each other:
┌────────────────────────────────────────────────────────┐
│ WORKFLOW SKILLS (25) │
│ Complete end-to-end PM processes (days to weeks) │
│ Example: run a full discovery cycle or write a PRD │
└────────────────────────────────────────────────────────┘
↓ orchestrates
┌────────────────────────────────────────────────────────┐
│ INTERACTIVE SKILLS (34) │
│ Guided discovery — 3-5 questions, then recommendations│
│ Example: "Which prioritization framework fits here?" │
└────────────────────────────────────────────────────────┘
↓ uses
┌────────────────────────────────────────────────────────┐
│ COMPONENT SKILLS (36) │
│ Templates for specific PM deliverables (30-90 min) │
│ Example: write a user story with acceptance criteria │
└────────────────────────────────────────────────────────┘
Interactive skills use an Adaptive Decision Ladder. Instead of dumping a framework at you, an interactive skill asks 3-5 targeted questions about your specific context, then offers numbered recommendations — each with a clear "use this when" rationale. You pick a path. The skill executes it and explains the why as it goes. If you want to just learn the framework without doing the work, you can ask that too — the skill coaches you either way. This is ABC — Always Be Coaching — in practice.
Full catalog: by type or by lifecycle bundle, all 95 skills with descriptions, or browse skills/ directly.
How a Skill File Works
Every SKILL.md follows the same structure:
| Section | What it contains |
|---|---|
| Frontmatter | name, description, type, intent, best_for, scenarios |
| Purpose | What this skill does and when to reach for it |
| Input | What you can bring (with example invocations) — inline input is used, not re-asked, and arriving empty-handed is fine: the skill walks you through it |
| Key Concepts | Frameworks, definitions, anti-patterns — with vocabulary explained |
| Application | Step-by-step instructions an agent (or human) can follow |
| Examples | Real-world cases showing both good and bad versions |
| Common Pitfalls | Named failure modes with consequences and corrections |
| References | Related skills and external frameworks |
The best_for frontmatter field lists 3-5 specific scenarios where the skill is most useful — helpful for quickly scanning whether a skill fits your situation.
Why no $ARGUMENTS templating? Other skill libraries use Claude Code's $ARGUMENTS substitution for input. We deliberately don't: it only expands in Claude Code (it renders as literal syntax in Claude Desktop/Web, Codex, and the Streamlit playground), and it teaches the human reader nothing. Instead, every skill has a plain-language ## Input section that works on every runtime — and makes clear you can show up with full context, partial context, or nothing at all and be guided through the rest. Full rationale in CONTRIBUTING.md.
Guided Context Capture. Every interactive skill gathers context the same way. You can go one question at a time, paste what you know, or say "best guess." A pasted dump gets read back to you as Known / Assumed / Missing / Conflicting before any question is asked. You get a recap before anything is generated. Decisions come as a ranked ladder (Recommended first, real alternatives with tradeoffs), and you make the call. The skill never re-asks what you already said or passes a guess off as a fact. Guided is the default opening; change it library-wide or per skill in pm-skills.config.yaml. How it works and why: docs/guided-context-capture.md.
Works With
Claude Code · Claude Desktop · Claude Web · OpenAI Codex · ChatGPT · Cursor · Windsurf · n8n · LangFlow · CrewAI · Gemini · OpenCode · any agent that reads structured markdown
See docs/Platform Guides for PMs.md for platform-specific setup.
Docs
| Document | Purpose |
|---|---|
| Using PM Skills 101 | Beginner-friendly orientation — setup without technical overload |
| Platform Guides for PMs | Tool-by-tool setup chooser for every supported platform |
| Using PM Skills with Claude | Claude Code + GitHub ZIP upload for Claude Desktop/Web |
| Using PM Skills with Codex | Local workspace + GitHub-connected Codex on ChatGPT |
| Using PM Skills with ChatGPT | GitHub app, Custom GPT Knowledge, and Project-based usage |
| Using PM Skills with Slash Commands 101 | Turn skills into reusable slash commands like /pm-story |
| Add-a-Skill Utility Guide | End-to-end guide for generating and validating new skills |
| Market Intelligence Suite Summary | The 14-skill competitive/market research suite: disciplines, chain, and which skill to run when |
| Lifecycle & EOL Suite Summary | The 8-skill lifecycle/sunset suite: the three plays, the EOL chain, and right-sizing |
| Building PM Skills | How raw PM content gets distilled into agent-ready skills |
| START_HERE.md | 60-second onboarding for local repo users |
What's New
v0.86 — November 13, 2026 · The Everyday Work of Product, Now on the Map
- 12 new skills fill the map: Discover gains Dig (
survey-design-advisor,usage-data-investigation,discovery-habits-advisor) and Decode (research-synthesis,assumption-mapping); Develop gains Drumbeat (backlog-refinement-advisor,sprint-planning,release-readiness); Drive gains Delineate (decision-rights-matrix,pm-role-charter); Deliver gains Deepen (activation-design,retention-diagnostic-advisor) - Pick your slice, not all 95: agents are converging on installable plugins and bundles, so the lifecycle bundles let you install what matches your work
- Easier to start: a five-minute path on the README's first screen, a layered
START_HERE.md, a clearer Desktop install, and a docs style guard - Older packs
02-07retired; every skill in them lives in a lifecycle bundle - Library at 95 skills · full story in the release note
v0.85 — November 13, 2026 · Lifecycle Bundles, a Navigator, and the Go-to-Market Gap Filled
- Six installable lifecycle bundles that follow the product arc: Determine, Discover, Define, Develop, Deliver, plus Drive (stakeholders and career) underneath. Install one with
/plugin install pm-define@pm-skills, or download its pack for Claude Desktop. Every skill still runs on its own pm-skills-navigator: describe what's on your desk and get routed to the right phase and skill- Go-to-market, filled:
launch-plan,launch-metrics,message-house,sales-value-narrative,buying-journey-enablement - Also:
LICENSING.md, an OpenCode install guide, and drift guards for prose counts - Library now at 83 skills · full story in the release note
v0.84 — August 10, 2026 · The Lifecycle & End-of-Life Suite
- 7 new skills + 1 upgrade for the part of product management nobody trains you on: what to do with a product that has stopped growing, and how to retire one without losing the customer
- Start upstream of the kill decision.
lifecycle-play-advisorandproduct-lifecycle-playsdiagnose where a product actually sits using seven transition questions, then pick between the three plays — extend, replace, or retire. Includes the seven replacement hazards and a risk register whose contingency column asks the question everyone skips: what is Plan B? - The EOL chain:
eol-readiness-advisor(go/no-go, and it will tell you to hold) →eol-stakeholder-sequence(Legal before Finance before Sales — get the order wrong and you find the landmines after the announcement) →eol-checklist(phase-gated, an owner on every item) →eol-internal-enablement(your teams ready before customers hear) →eol-message(upgraded: three transition paths including the honest no-replacement case) →eol-process(the whole thing, six phases, including the post-EOL review everybody skips) - Right-sizing is a dial you set, not a verdict handed to you. Every skill offers Light / Standard / Heavy, defaults to Standard, and names what a lighter choice leaves uncovered. A feature deprecation should never get a flagship playbook — that's how teams learn to ignore EOL process on the sunset that actually matters
- No tight coupling. Every skill runs standalone; the workflow is a recommended route through independent stops, not a pipeline. Enter wherever you actually are — including mid-sunset, with a diagnostic that maps symptoms ("Support is drowning") to the phase that got skipped
- Library now at 77 skills · full story in the release note
v0.83 — July 17, 2026 · The Market Intelligence Suite
- 14 new skills + 2 major upgrades — the biggest single addition in the library's history. Run competitive and market research like an intelligence agency instead of a term paper: eight collection disciplines (OSINT → MASINT), signal → inference chains, confidence stacking, and a fusion cadence
- New interaction mode:
autonomous-investigation— the protocol for research that proceeds without you. Question budgets, a search-plan gate, Fact / Inference / Assumption labels on every claim, do-not-invent lists, and stable diffable schemas, so investigations can run on a schedule and you can diff this quarter against last - The investigation chain (market-landscape-scan → competitive-research-snapshot → competitive-intel-watch → battle-card-builder), evidence-cited strategy frameworks (swot-analysis, porters-five-forces, ansoff-matrix), monitors and miners (voice-of-customer-miner, pricing-packaging-tracker, pestel-delta-monitor), and the six-step umbrella competitive-analysis-process
- Upgraded:
tam-sam-som-calculator(three entry modes, including autonomous bottom-up research) andcompany-intel(Executive Signal Refresh rerun pattern — Then/Now diffs and Dropped Language: what leaders stop saying is often the strongest signal) - Library now at 70 skills · full story in the release note
v0.82 — July 8, 2026
- Added
incoming-request-advisor(Interactive) — drop in a Slack ping, email, mandate, or escalation and get a structured breakdown that separates the literal ask from the real job-to-be-done, reads sender power and stake, and points you toward a reply. Ships with a copy/paste template so you can run it by hand too - New: a browsable download shelf at
/dist— no terminal, no Releases tab. Read the plain-language README, scan the CATALOG, and download any skill or pack straight from the repo. Built for PMs who just want the skills - Library now at 56 skills
v0.81 — July 4, 2026
- Every skill now has a required
## Inputsection: what to bring, what happens to context you supply up front (it's used, not re-asked), and reassurance that arriving empty-handed is fine — the guided flow covers the rest - Added
argument-hintautocomplete for Claude Code users; deliberately no$ARGUMENTStemplating — it breaks on every other runtime and teaches the reader nothing (why) - Validator now enforces the convention: skills fail without an Input section or with bare
$ARGUMENTSin the body - Streamlit playground shows each skill's "What to bring (all optional)" before you start a session
- Restored
agent-orchestration-advisor(Interactive) — the multi-agent workflow design skill was referenced everywhere but only existed on an orphaned commit; recovered from git history and brought up to current standards
v0.80 — June 19, 2026
- Added
stakeholder-identification(Component) — comprehensive stakeholder brainstorm using allies/audiences/influencers, R/P/D marking, equity lens, and bias check; narrows to priority targets - Added
stakeholder-mapping(Component) — two complementary grids (Power × Interest + Impact × Power); comparing outputs reveals who you're under-engaging relative to how much the product affects them - Added
stakeholder-engagement-advisor(Interactive) — per-stakeholder engagement planning via Adaptive Decision Ladder: three questions on profile, power/impact, and context deliver tailored message framing, medium, cadence, and a named next action
All three adapted from the MITRE Innovation Toolkit via the companion repo MITRE ITK Skills — worth a bookmark if you work in discovery, facilitation, or cross-functional product strategy.
v0.79 — May 15, 2026
- Added
organic-growth-advisor— McKinsey Growth Pyramid triage for new segments, geographies, channels, or products - Added
pm-skill-creator— interactive skill for designing repo-compliant skills via guided conversation - Fixed missing
.claude-plugin/plugin.jsonthat silently blocked Claude Code skill discovery - Added configurable input length guard (
PM_MAX_INPUT) and path traversal protection to helper scripts
Contributing
Found a gap? Have a PM framework worth formalizing? The bar is pedagogic — skills must teach the why, not just the how.
See CONTRIBUTING.md for guidelines, or open an issue to start a conversation.
License
CC BY-NC-SA 4.0 — non-commercial use with share-alike.
Everything in this repository — every skill, template, and doc — is licensed CC BY-NC-SA 4.0. There is no mix of licenses here.
Some skills note in their Provenance sections that they were adapted from product-manager-prompts, Dean's earlier prompt library. That repo has the same author, so there is no license conflict: a license grants permissions to other people, and a copyright holder is free to adapt and relicense their own work. Those Provenance lines are lineage — a breadcrumb back to where an idea started — not a license dependency. No third-party MIT-licensed text is incorporated anywhere in this library.
In plain terms:
- ✅ Use these skills in your day job — at a for-profit company, with your team, in your agents. That's what they're for.
- ✅ Adapt and remix them — share what you build under this same license, with credit.
- ✅ Teach with them — workshops, brown bags, mentoring, sending the ladder down.
- ❌ Don't sell them — no repackaging the skills themselves into a paid product, course, or service without expressed written permission.
- 📄 Need something citable for your legal team? See LICENSING.md: internal use at a for-profit company is fine, and what you produce using the skills is yours.
- 🤔 Not sure your use qualifies? Open an issue and ask. If you're using these in the spirit they were built — to get better at the craft and help others do the same — the answer is almost certainly yes.
The companion prompt library, product-manager-prompts, carries the same CC BY-NC-SA 4.0 license as of its v2.3.0, with its own plain-language permissions (stricter on commercial use) — see its LICENSING.md, which governs that repo.
Questions
- GitHub Issues: Report bugs or suggest features
- LinkedIn: Dean Peters
- Productside: AI PM consulting and training
Details
- Marketplace
- deanpeters/Product-Manager-Skills
- Source
- skills/retention-diagnostic-advisor
- Category
- Delivery
More from deanpeters/Product-Manager-Skills
All 102- roadmap-planning
- saas-economics-efficiency-metrics
- saas-revenue-growth-metrics
- sales-value-narrative
- skill-authoring-workflow
- sprint-planning
- stakeholder-engagement-advisor
- stakeholder-identification
- stakeholder-mapping
- storyboard
- survey-design-advisor
- swot-analysis
- tam-sam-som-calculator
- usage-data-investigation
- user-story
- user-story-mapping
- user-story-mapping-workshop
- user-story-splitting
- voice-of-customer-miner
- vp-cpo-readiness-advisor
- workshop-facilitation
- acquisition-channel-advisor
- activation-design
- agent-orchestration-advisor
Similar to retention-diagnostic-advisor
More Development Tools
reamorluto/rea466.8K starsReverse engineer anything with agents, from app behavior down to native binaries.
superpowersobra/superpowers319.5K starsCore skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques
claude-apianthropics/skills196.9K starsClaude API and SDK documentation skill for building LLM-powered applications
prompts.chatf/prompts.chat178.2K starsAccess thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
agent-sdk-devanthropics/claude-code156.1K starsDevelopment kit for working with the Claude Agent SDK
ui-ux-pro-maxnextlevelbuilder/ui-ux-pro-max-skill152.6K starsProfessional UI/UX design intelligence for AI coding assistants. Includes searchable databases of styles, colors, typography, charts, and UX guidelines for Reac