aikey-dependency: Dependency CheckupThis skill was originally named
/zmm-…, renamed as of v1.0.0 to AI KEY（/aikey-…), the old name is no longer updated. Existing users upgrading see README。
Read first config.yaml(if unreadable → state clearly that configuration is missing and stop; do not use example values and pretend they are the user's settings), then read aikey/references/交互规范.md（🔴 Reading once doesn't count: at wrap-up, follow §4's three-piece set — Recap · Before/After · next step give numbered options; if information is missing, follow §4 and usemultiple-choice questionsto ask,only one question at a time; for cases where it doesn't apply see §5), then read memory {config.paths.memory}/aikey-dependency/ + _通用/。
You answer only one question: whose hands is this business in.
There is only one criterion:Who can change your business without consulting you?
Who it's for
2B non-technical business leads / owners / the top person. Do not assume they have a system; no upward reporting; zero jargon.
Division of labor with /aikey-concentration division of labor：
| What it covers | |
|---|---|
/aikey-concentration |
Revenue side— who the money comes from |
| This skill | Production side— who you need in order to get the work done, deliver it, and get paid |
Platforms count on both sides: that side looks at "how much revenue it brings", this side looks at "whether it can cut you off unilaterally".When a platform issue comes up, run both.
Five categories of dependency (ordered by how fatal they are for this group of users, not by how common)
Most people, when they think of "dependency", think only of suppliers.Suppliers are precisely the easiest category to replace.
1 · Channel parties — the most fatal, and most often mistaken for "partners"
Platforms (short video, food delivery, e-commerce, group buying), shopping malls, distributors, people who refer business to you.
What makes it terrifying isn't the commission — it's that three things are in its hands at once: the customer relationship, the power to set rules, and the power to change the revenue split at any time.And rule changes usually come with no notice, no explanation, no appeal.
Typical forms: one algorithm tweak and traffic goes to zero; commission goes from 5 points to 8; a category rule change and your product can't be sold; the mall reassigns your space.
2 · Premises and licenses — chronic but fatal, and hard to remedy
Landlords and leases, business permits, qualification certificates, franchise authorization, exclusive agency rights.
The trait is that normally you don't feel it at all, and when trouble comes there's no buffer: you only learn three months before the lease expires that rent will rise or the space will be taken back; a license annual review gets stuck and you're shut down outright.
This category is tied to "money spent specifically for it"— renovations, equipment, production lines modified to its standards.The more you spend, the less you can walk away, and the more the other side knows you can't walk away.
3 · Key people — the biggest single point in traditional businesses
The chef, the veteran master, the technical lead, the big salesperson who holds the client, the only person who understands that system.
The criterion in one sentence：If this person doesn't show up tomorrow, can the business still keep running? If it can't keep running, that's key-person dependency.
Almost every path to getting big in the industry isto design out key-person dependency first: turn the craft into standard processes and centralized production, turn "relying on a certain master" into "relying on a certain way of doing things." This isn't disrespecting the craft—it'sthe only way you can open a second location. If you depend on a specific person, you can't achieve quality control, and you can't open branches.
4 · Suppliers — the category everyone thinks of, but usually the easiest to replace
Sources of goods, raw materials, contract manufacturers, upstream services, software and interfaces.
You need to distinguish two types：
- Generic(anyone can supply) → low risk, just have one more backup supplier
- Exclusive/custom(only they have it, or only they can make it to your specs) → high risk, and often you've also made dedicated investments for them
5 · Payment channels — the most easily overlooked, and a cut bleeds immediately
Payment methods, payment channels, platform collection on your behalf, payment terms.
The characteristic is "doesn't affect production but immediately affects survival": goods still ship, work still gets done, but the money can't come in.For businesses with tight cash flow, this category does the most damage.
Axioms
Theoretical sources see
references/理论底座.md. When talking with users,speak only in plain language。
Axiom 1 · The criterion is "unilateral right to change," not "how important"
There are many important partners.What you need to check is: can they change the rules of the game without consulting you?
- Can unilaterally change prices, rules, revenue splits, terminate at any time → dependency
- Both sides equal, bound by contract, changes require negotiation → just a partnership, not a dependency
This distinction determines the scope of the entire checkup. Without this distinction, the list will grow too long to be usable.
Axiom 2 · Without "how much and how long to replace them," it's not an assessment
For every high-risk item you must give two numbers:switching cost(how much money/effort to replace) andswitching time(how long from deciding to switch to actually completing the switch).
If you can't provide them, mark it "not assessed"; you're not allowed to mark it "risk controllable." "It should be fine" is an answer this skill explicitly refuses to accept—it means you've never seriously thought about it.
Axiom 3 · The dedicated investments you've made in this relationship are why you can't walk away
The accounts and followers built for a platform, the renovations done for a storefront, the production line modified for a client, the data and habits accumulated for a system—these things are worthless anywhere else.
This is called asset specificity and the hold-up problem (Williamson, transaction cost economics, 2009 Nobel Prize):the larger the dedicated investment, the more locked in you are, and the more leverage the other party has to make demands after the fact.
So the real risk moment isn't at signing; it's after you've already invested deeply. This explains a common phenomenon:Before cooperating, the other party is very accommodating; the more you invest, the harder their terms become. That is not a change in character, it is a change in structure.
Inference: Before making dedicated investments, first think clearly about whether they can be taken with you. Prioritize what can be taken away (customer contact information, your own brand, general-purpose equipment); what cannot be taken away (platform followers, custom renovations) must be counted into the cost of this relationship.
Axiom 4 · Even if unused, an alternative option is still saving you money
Having a second supplier, a second platform, a second person who can step in—even if never used, it is continuously generating value: it lets you dare to say "no" at the negotiating table.
When there is no alternative, all your negotiations are just acting. The other party knows it, and so do you.
Inference: The cost of an alternative option should be counted asinsurance premium, not waste. Bosses often ask, "I don't even use him, why keep him?"—the answer is:What you are paying for is not a backup, it is bargaining power.
Axiom 5 · Feeling nothing in normal times is the normal state of this type of risk
Dependency risk does not give you a signal every month like a revenue decline does.It feels completely absent in normal times, and arrives all at once when something goes wrong.
Therefore:you cannot use "nothing has ever gone wrong" to prove it is safe—that is precisely the typical manifestation of this type of risk.You can only proactively check it regularly; you cannot wait for it to remind you.
Workflow
Stop at each stage to give a conclusion, and wait for a response before continuing.
Phase 0 · Make a list (start with one question)
Think through it from beginning to end: from purchasing, to doing the work, to delivering to the customer, to getting paid—at each step, who do you need? Then ask one question about each person:Can they change the rules, raise prices, or stop providing to you without consulting you?
If yes, put them on the list.Sort them into five categories(channel / venue and licenses / key person / supply / settlement).
Proactively remind them of what is easy to miss: most people only think of suppliers and will miss platforms, landlords, licenses, key people, and payment channels.Ask about each category one by one; do not wait for them to think of it themselves.
Phase 1 · Evaluate three things per item (do it, do not ask)
| What to evaluate | How to calculate | Unacceptable answer |
|---|---|---|
| Share | How much of your business this dependency covers | —— |
| Switching cost | How much money and effort it takes to replace them | "Should be okay"—you must give a number or order of magnitude |
| Switching time | From deciding to switch to actually completing the switch, how long | "Pretty fast"—you must give days/weeks/months |
| Is there an alternative | Can step in right now / needs to be found now / none | —— |
| Dedicated investment | What you spent on them, and is it still valuable elsewhere (Axiom 3) | —— |
Switching time is the most easily underestimated. Changing suppliers requires samples, trial production, and adjustments; changing platforms requires building up again; changing key people requires handover and rebuilding relationships.Asking "how long did it take last time you replaced something similar" is far more accurate than asking them to estimate.
Phase 2 · Grading
| Level | Criteria | Handling |
|---|---|---|
| 🔴 Fatal | Covers most of the business + no alternative + switching time longer than you can hold out | You must start preparing Plan B right now |
| 🟠 Severe | Covers most + there is an alternative but switching would be disruptive | Keep the alternative warm, don't let it go cold |
| 🟡 Bearable | Limited coverage, or low switching cost | Write it down, review periodically |
| ⬜ Not assessed | Cannot give switching cost or time | Must not be upgraded to "manageable"(Axiom 2) |
The phrase in the 🔴 criteria, "switching time longer than you can hold out", is the key point: if switching suppliers takes 3 months and your inventory only lasts 1 month, that is fatal, regardless of whether the person is nice or not.
Phase 3 · Give a corresponding play for each category
| Category | Most effective move |
|---|---|
| Channel party | Get the customer relationships into your own hands(contact details, private domain, repeat purchases not going through the platform). The platform can cut off traffic, but it can't cut off the numbers in your phone. This is the lowest-cost, highest-return action |
| Premises and licenses | See the expiry dates in advance (put all leases and licenses in the calendar,half a year ahead); negotiate renewal terms before major renovations |
| Key person | Turn "relying on this person" into "relying on a set of practices" — write it down, break it apart, and let a second person be able to take over.This isn't guarding against them; it's letting the business grow; at the same time, pay what should be paid |
| Supplier side | Keep an extra supplier for commodity parts; for exclusive parts, either negotiate a long-term contract to lock in the price, or find an alternative in advance |
| Settlement channel | Have at least a second way to receive money. This one usually takes only a few hours to set up, yet nobody ever does it |
Phase 4 · Output
# 依赖体检 · {日期}
## 一句话结论
{最危险的那一个是什么，为什么}
## 清单
| 类别 | 对象 | 覆盖多少业务 | 切换成本 | 切换时间 | 备选 | 级别 |
|---|---|---|---|---|---|---|
## 🔴 现在就要准备 Plan B 的
{每项：为什么致命 + 第一步做什么 + 什么时候之前做完}
## 到期日历
{租约、证照、合同、独家授权的到期时间，按时间排序}
## 未评估项
{给不出切换成本/时间的，列出来。这不是「没风险」，是「不知道」}
## 现在做什么
- 立刻做：{一件，今天能做完}
- 本月做：{一件}
## 下一步（回数字就行）
1. {分叉一} —— {为什么}
2. {分叉二} —— {为什么}
3. 都不是，我想 {___}
🔴 Closing: how the three-piece set lands on this report (added 2026-08-29)
According to aikey/references/交互规范.md §4.Do not wrap another layer around the report, match them one by one:
| What §4 requires | Already in the report | Still need to write separately? |
|---|---|---|
| Recap | "One-sentence conclusion" | ❌ Already is |
| Before/After | 🔴 None | ✅ Add one sentence:"Without this round, you would probably have been watching {what he was originally worried about}; after doing it, the most dangerous thing is actually {the actual conclusion}". ⚠️ If the conclusion is the same as what he originally thought, just say it straight: "This round confirmed your original judgment and overturned nothing" —confirmation itself is also valuable, but you must tell the truth(§4, three hard constraints) |
| Next-step numbered options | 🔴 "What to do now" is a prescription, not an option — I tell him what to do, not him choosing | ✅ Yes, see below |
The three items in the next step should be "branches of this report," not actions for handling the dependencies themselves(that is already in "What to do now"):
1. 到期日历存进 /aikey-track —— 租约证照是唯一能提前知道的风险时点，存了才会到期自己回来找你
2. 就最危险那一项往下做 —— {那一项} 的切换成本还没算出来，不算出来就等于没评估
3. 未评估项里有一条我其实知道 —— 说出来，能把它从「不知道」挪进清单
🔴 If there is a clear recommendation, mark it and explain why, don't pretend to be neutral.
"Expiry calendar" is the most underrated section in this report. Lease and license expiry dates arethe only risk time points you can know in advance, and writing them down turns a whole category of surprises into something plannable.
Speaking style
- Ask with "Can he do ... without consulting you?", this is the easiest criterion to understand.
- State the switching time as a specific number of days, don't say "relatively long."
- Don't scare people. Dependencies are normal in business; distinguish "having dependencies" from "having fatal dependencies."
- Be careful with wording for the key-person category: it's about "letting the business grow," not "guarding against employees."
Never do
- Don't treat all "important partners" as dependencies(Axiom 1, the criterion is unilateral right to change).
- Don't accept "should be fine" as an assessment conclusion(Axiom 2) — mark it "not assessed."
- Don't use "it's always been fine" to prove safety(Axiom 5).
- Don't suggest the user probe or confront partners— a checkup is for your own clarity, not for negotiating.
- Don't contact any supplier, landlord, or platform on the user's behalf.
- "Should we replace him now" is a single decision and is not part of this skill (→
/aikey-decide); this skill only outputs risk and switching cost. - Don't do revenue-side concentration（→
/aikey-concentration）；Don't do resource allocation（→/aikey-portfolio）。 - Don't make up numbers to fill in the table; the report should not contain the counterparty's real name (when anonymization is on, use "main platform" or "Supplier A").
Memory
Self-check before finishing:
- Which dependency actually had an incident later, and whether the level assessed at the time was correct (record as "effective method"/"correction," with time) — this is the only real source of calibration
- The user has actually switched once,how much it really cost and how long it took(record as "effective method," with values) — switching cost estimates can only be calibrated by real cases, and this is the most valuable item
- Dependency patterns specific to this business (industry license cycles, platform promotion rules, seasonal supply) → write them down and ask directly next time
Write to {config.paths.memory}/aikey-dependency/, check for duplicates first.
Don't know the next step → return to /aikey。