メニュー
Run a go/no-go assessment for retiring a product or feature, then right-size the effort. Use when someone says "we should probably kill this" and nobody has made the call.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/deanpeters/Product-Manager-Skills --skill eol-readiness-advisorコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Understand the PM-to-Director transition through altitude and horizon thinking. Use when diagnosing scope, time-horizon, or leadership-level gaps.
Create a company research brief with executive quotes, product strategy, and org context. Use when preparing for interviews, competitive analysis, partnerships, or market-entry work.
Plan a VP or CPO 30-60-90 day diagnostic onboarding path. Use when entering a new executive product role and avoiding premature change.
SOC 職業分類に基づく
| name | eol-readiness-advisor |
| argument-hint | [product or feature under EOL consideration, and what triggered it] |
| description | Run a go/no-go assessment for retiring a product or feature, then right-size the effort. Use when someone says "we should probably kill this" and nobody has made the call. |
| intent | Structured go/no-go assessment for sunsetting a product or feature. Determines whether EOL is the right call, how intense the sunset effort needs to be, and which organizational functions need to be involved. Prevents both premature kills and expensive delays on products that should have been retired sooner. |
| type | interactive |
| theme | eol-transition |
| best_for | ["Deciding whether a declining product should actually be retired — or held, harvested, or fixed","Right-sizing an EOL effort so a feature deprecation doesn't get a flagship playbook","Naming the obligations and landmines that make a sunset harder than it looks"] |
| scenarios | ["Leadership keeps saying we should kill this module but nobody will make the call — help me assess it","We're retiring a hardware line with service contracts and channel partners; how big does this effort need to be?"] |
| estimated_time | 15-25 min |
Decide whether a product or feature should actually be retired — and if so, how much machinery the retirement deserves. This skill produces a go/no-go verdict with named evidence, an intensity level you choose, and a short list of the obligations that will bite you if ignored.
Most EOL guidance assumes the decision is already made and jumps to the plan. This skill sits earlier: it is the conversation where "we should probably kill this" becomes a decision someone will defend in a room.
It is equally willing to tell you not to retire. A "hold" or "harvest" verdict is a real outcome here, not a consolation prize.
Works best with: The product or feature under consideration and what triggered the conversation (a declining number, a strategy shift, a support-cost complaint, an exec offhand remark).
Also useful: Revenue and customer counts, whether a replacement exists, contract or regulatory commitments, and any political sensitivities you already know about.
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 opens by asking what's under review and why, then walks you through blast radius, intensity, and transition path one question at a time. You do not need the numbers in front of you — "I don't know" is a valid answer that becomes a labeled assumption in the verdict.
Example invocations:
Should we retire our legacy reporting module? Usage is down 60% and support costs are climbing.EOL readiness check on our NFA-200 controller line — 120 installs, channel partners, UL certified.Not all EOLs play out the same. Some are a changelog entry and a support macro. Some consume the whole company for three quarters. Most land in the middle. The single most common EOL failure in either direction is a mismatch between the size of the sunset and the size of the response — a flagship retirement run as a Jira ticket, or a feature deprecation that spawns a steering committee.
So the intensity is a dial you set, not a verdict the framework hands you:
| Dimension | Level 1 — Light | Level 2 — Standard | Level 3 — Heavy |
|---|
| Typical scope | Feature, internal tool, unversioned API | Commercial product with active customers | Revenue-critical, hardware, regulated |
| Stakeholder stops | 3-4 | 7-8 | 10+ |
| Lifecycle gates used | 2-3 (NSC, EOS, EOL) | 4-5 (NSC, EOS, EOE, EOM, EOL) | All 8 (GA through EOSRV) |
| Functional areas | Product, Eng, Support, Docs | + Sales, Marketing, CS, Finance, Legal, IT, Data | + Supply Chain, Channel, Regulatory, Org |
| Enablement | Support FAQ only | + Sales talking points, objections, escalation | + Channel brief, account tiers, training |
| Customer message | Brief notice | Standard with phase table | Full with compliance section |
| Typical lead time | Weeks | 6-12 months | 12-24 months |
Level 2 is where most sunsets belong. Treat it as the default and argue your way up or down.
The dial is yours to move. Recommend a level, explain the reasoning, then let the user override in either direction — and let them move it again later. A PM who says "give me the light version, I know my org" is exercising judgment, not making an error. Honor it, and note in one line what the lighter level leaves uncovered so the choice is informed rather than blind.
EOL is not one date. It is a series of gates, and most customer anger comes from collapsing them into a single announcement:
Light sunsets use three of these. Heavy sunsets use all eight. Naming the gates you're not using is as useful as naming the ones you are.
Evidence that supports a go verdict. Two or more, strongly present, is a real case:
Evidence that argues against retiring now, even when the retire signals are loud:
Use workshop-facilitation as the default interaction
protocol for this skill.
This skill asks up to 4 adaptive questions, then delivers a verdict, an intensity level, and numbered next steps. Give the heads-up at the start: four questions, roughly fifteen minutes, and they can dump context at any point to skip ahead.
Agent asks:
"What product or feature is under EOL consideration, and what brought it up? The trigger matters as much as the product — a declining metric, a strategy shift, a support-cost complaint, and an exec offhand remark lead to very different assessments."
Is retirement even the right question yet? This skill assesses one play: retiring. At the mature-to-decline inflection there are three — extend the product, replace it with a successor, or retire it. If the user hasn't established that retirement is the play, say so and offer the upstream diagnosis:
"Before we assess retiring it — is retiring the play you've settled on? There are two others (extend it, or replace it with a successor), and
lifecycle-play-advisorworks out which fits. Happy to continue here if the call is made."
Don't force the detour. A user who says "we've decided, assess the retirement" gets the assessment.
Listen for the trigger type, because it predicts the failure mode:
| Trigger | Watch for |
|---|---|
| Declining metric | Is it decline, or seasonality/measurement change? |
| Strategy shift | Is the product misaligned, or just unfashionable? |
| Support cost | Is retirement the fix, or is a bug fix the fix? |
| Exec remark | Is there evidence, or is this an opinion with a title? |
| Replacement shipped | Is the replacement actually at parity for these customers? |
If the trigger is an exec remark with no supporting evidence, say so plainly at verdict time. That is the single most common source of premature kills.
Agent asks:
"Help me size the impact. Rough numbers are fine, and 'I don't know' is a real answer:
Why this matters (say it): These four answers set the intensity level. Item 3 is the one that most often turns a "simple" sunset into a legal problem — it is asked early on purpose.
Mark any unknown as an assumption to be labeled in the verdict. Do not stall the flow to chase a number; an assessment with three labeled assumptions beats an assessment that never happened.
Agent presents the recommendation, then hands over the dial:
"Based on what you've described, here's my read on how much machinery this deserves:
Recommended: Level [N] — [Light / Standard / Heavy] Because: [two or three specifics drawn from their answers]
Take the recommendation, or pick a different level — you know your organization. You can also change it later; nothing downstream is locked to this choice."
If the user dials down, accept it and add one line: "Going lighter than recommended means [the specific thing that goes uncovered — usually Legal review, channel notification, or migration support]. Flagging it so it's a choice, not a surprise."
If the user dials up, accept it without argument. Over-preparing an EOL is a cheap mistake.
Agent asks:
"What happens to the people using it? Three broad paths:
Which is it, and how ready is that path today?"
Path readiness is a go/no-go input, not a detail. A replacement that is not at feature parity for these specific customers is not a landing place yet. If the path is unready, the verdict is usually "go, but the announcement waits until [specific readiness condition]."
Produce this structure:
## EOL Readiness Assessment: [Product]
**Verdict:** [Go / Go-with-conditions / Hold / Harvest]
**Intensity:** Level [N] — [Light/Standard/Heavy] *(user-selected)*
**Confidence:** [High/Medium/Low] — based on [what's known vs. assumed]
### The Case to Retire
- [Retire signal + the evidence behind it]
### The Case Against (or: Conditions)
- [Hold signal + the evidence behind it]
### Assumptions I Made
- [Anything answered "I don't know," stated plainly]
### Obligations to Check Before Announcing
- [Contracts, SLAs, regulatory, "lifetime" promises, channel agreements]
### Landing Place
- **Path:** [Replacement / Migration / Graceful exit]
- **Readiness:** [Ready / Gap: specific gap / Not started]
### What This Level Covers — and Doesn't
- Covers: [gates, stakeholder count, artifacts implied by the level]
- Does not cover: [what the level deliberately skips]
Then offer numbered next steps:
"Where do you want to go next?
eol-messagePick a number, combine them ('2 & 4'), or tell me what you'd rather do."
On the verdict names:
lifecycle-play-advisor to size the playHold and Extend are different findings. Hold says not now; Extend says this product still has a move available. Don't collapse them — a PM told "hold" waits, and a PM told "extend" acts.
PM: "Should we kill Fieldlight Classic? It's our legacy dispatch module. Leadership has been hinting at it for two quarters."
Agent (Q1): "What triggered the hints — a metric, a strategy shift, or a cost complaint?"
PM: "Support tickets. Classic generates about 40% of our support volume on maybe 15% of revenue. And we shipped Fieldlight Next last year."
Agent (Q2): "Blast radius — customers, revenue, commitments, and any channel or hardware?"
PM: "About 800 accounts, $2.4M ARR. Annual contracts, standard terms, nothing regulatory. Pure SaaS, we sell direct."
Agent (Q3): "Recommended: Level 2 — Standard. 800 accounts with annual contracts and real revenue means phased comms and a cross-functional checklist, but no regulatory or channel workstream, so you don't need the heavy version. [presents the three levels]"
PM: "Honestly we're a 40-person company. Give me Level 1."
Agent: "Taking Level 1. Going lighter here means you skip the formal Legal review of the renewal-date language and the phased sales enablement — with 800 annual contracts, the renewal dates are the piece I'd least like to see uncovered. Flagging it so it's a choice, not a surprise. Want to keep Level 1 with a single Legal check bolted on? That's a common middle."
PM: "Yeah, do that."
Agent (Q4): "Where do the 800 accounts land — Fieldlight Next, or something else?"
PM: "Next. It's at parity except for the custom dispatch rules maybe 30 accounts use."
Verdict: Go-with-conditions. Level 1 + Legal check. Condition: the 30 custom-rule accounts need a named plan before announcement, or they become your escalation queue. Assumption labeled: support volume figure is from memory, not pulled.
PM: "We want to retire the NFA-200 retrofit controller. The NFA-500 launched and manufacturing wants the line freed up."
Agent (Q1): "So the trigger is internal — manufacturing capacity plus a shipped successor. Any customer-side signal, or is this entirely supply-side?"
PM: "Entirely supply-side. Customers are fine with the 200."
Agent (Q2): "Blast radius — installs, revenue, commitments, hardware and channel?"
PM: "About 120 installations. Eight channel partners resell it. Service contracts run through 2028 on some. UL and CE certified. And there's spare parts inventory."
Agent (Q3): "Recommended: Level 3 — Heavy. Service contracts running to 2028, eight channel partners, regulatory certification, and physical spare parts — that's four separate workstreams the lighter levels don't have room for."
PM: "Agreed, Level 3."
Agent (Q4): "Where do the 120 installs land? Is the NFA-500 a drop-in for the 200's installed base?"
PM: "Not exactly. The 500 needs a different mounting bracket and the retrofit cases are... it'd be a site visit each."
Verdict: Hold. The retire signals are real but supply-side only. Against them: service contracts running to 2028 (obligation lock), and no drop-in landing place — 120 site visits is not a migration path, it's a project needing its own funding. Customers are not asking for this.
Recommended instead: EOS now, EOL later. Stop selling the NFA-200, keep servicing it through contract expiry, and fund the bracket/retrofit work as the actual prerequisite. Revisit when the retrofit path exists or contracts wind down — whichever comes first. Manufacturing gets the line back at EOS, which was the real ask.
Note how the verdict separates the presenting problem (retire the product) from the underlying one (free the manufacturing line) — and solves the second more cheaply.
Symptom: The entire case rests on an executive saying "we should kill that thing."
Consequence: You spend two quarters retiring a product that was quietly profitable, and nobody remembers whose idea it was when the churn shows up.
Fix: Name the trigger explicitly in the assessment. If the evidence is one remark, the verdict is "Hold — pending evidence," not "Go."
Symptom: Every EOL gets the same treatment — either everything runs as a ticket, or everything gets the steering committee.
Consequence: Light sunsets burn political capital on ceremony. Heavy sunsets blow up in public.
Fix: Set the intensity deliberately, say it out loud, and name what the chosen level skips.
Symptom: Blast radius is measured in customers and revenue; contracts and certifications go unasked.
Consequence: The worst EOL failures on record are legal, not commercial — refund orders over "lifetime" hardware promises, fines over retired services with no adequate alternative.
Fix: Ask about commitments in the same breath as revenue. It's question 2 for a reason.
Symptom: The verdict is Go, and the landing place is "customers will find something."
Consequence: You are not retiring a product, you are evicting customers. They remember.
Fix: Graceful exit is a legitimate path — but it requires data export, extended notice, and honest alternatives. If none exist yet, the verdict is Go-with-conditions.
Symptom: Someone asks to retire a product; the assessment evaluates retiring the product.
Consequence: You run a full sunset when an End-of-Sale, a price increase, or a bug fix would have satisfied the actual need.
Fix: Ask what the trigger is really about. "Free up the manufacturing line" is answered by EOS, not EOL. Name the cheaper answer when you see it.
Each of these stands on its own — none require this skill to have run first, and this skill is not a prerequisite for any of them. If you did run this assessment, carry your intensity level over by simply saying "Level 2" when you start; there's no handoff format to preserve.
eol-message — the customer-facing announcementincoming-request-advisor — when the EOL request arrived
as a message you need to decode firststakeholder-map — general stakeholder mappingbusiness-health-diagnostic — when the trigger is
financial and you want the numbers examined properlyworkshop-facilitation — the interaction protocolprompts/eol-readiness-assessment.md in the
https://github.com/deanpeters/product-manager-prompts repo.