Picks up where the opening pitch ends. Classifies the reply with a three-token test, then writes the four branches: the yes, the silence, the no, and the unpaid invoice.
/plugin marketplace add mkhalid1/locul-skills
Then /plugin to install Copywriting, which includes this skill.
We have not measured this one. It is published untested, and the fair assumption is that a strong model asked for a partnership email will write a decent opening message without any help at all.
The opening is not what this file is about. It picks up at the first reply, and it carries three things a general instruction does not. The first is a classification test that runs before any writing: count whether the reply contains a date, a name or a number, treat warmth as worth nothing, and read a warm token-free reply as ambiguous rather than as agreement. That single rule is aimed at the most expensive error in partnership work, which is a maybe filed as a yes and then reported as pipeline for a quarter. The second is the post-rejection message, given its own section with its structure and, more usefully, the reason it works: the decision has already been made, so nothing is left to protect in the negotiation and the honest answer costs the other side nothing, which is why it returns better material than a survey of people with nothing at stake. The third is the unpaid invoice, which is a commercial conversation in a completely different register from the other three and which most people handle badly because they are embarrassed, so it arrives either apologetic or, three weeks late, wrapped in legal vocabulary.
Where this file starts, because a reader needs to know which half of the job it is. It starts at the first reply. The opening approach to a stranger sits outside it: the research gate, the subject line, the escalating ask and the touches that chase a first answer are not here, and nothing in this directory carries that opening sequence end to end. If nobody has replied to you yet, this is not yet the file. It assumes a reply exists and a counterparty with their own assets, their own audience and their own approval chain, and it runs from that reply to a signature or to a recorded loss reason, including the paperwork chase and the money conversation that follow a yes.
Who it is not for. If your partnerships close in one conversation over a drink, the branch machinery is overhead. If you are running hundreds of small integrations through a self-serve directory, this is the wrong shape entirely and a programme with published terms is the right one.
Measuring one skill honestly costs about twenty model sessions: five runs with it, five without, on real material, each output graded alone by a session that is not told the other arm exists, against a rubric written by somebody who never saw the skill. We have not spent that on this one yet, so it ships labelled rather than ships silently.
How it would be measured. Tier A with a preference layer on top. Both arms receive the same six invented partnership threads, each carrying an opening message already sent plus one scripted reply: two commitments, two warm replies containing no date and no name and no number, one flat refusal, and one thread where an agreed invoice is nineteen days past its date. The mechanical grade asks whether the two token-free replies are classified as ambiguous rather than as agreement, whether the reply to the refusal contains zero product claims and exactly one question and one named month, whether every chase names a specific return time instead of a courtesy phrase, whether the invoice message names an amount and a window and a stated consequence with no apology, and whether a verbatim loss reason and a next dated action are recorded for every thread. Register and warmth have no ground truth, so the rest needs blind pairwise preference from judges who have run partnership deals.
Half of it grades without judgement: whether a token-free reply is classified as ambiguous, whether the rejection message contains zero product claims and exactly one question, whether a chase names a return time, whether the invoice message carries an amount and a window and a consequence. Those are lookups on the output.
The expensive part is the fixture and the ceiling. A fair test needs six invented threads with an opening message, a scripted reply and enough surrounding context that both arms are writing to a real situation rather than inventing one, which is a day of writing before a single run. And the honest expectation is that a strong model already writes a serviceable acceptance message and a serviceable chase, so if a gap exists it sits in the two branches people avoid, the refusal and the money, where there is the least training material because almost nobody publishes theirs.
The rule that decides pass or fail was written down before any run was executed and it does not move afterwards. It is in the method note on the hub, along with the full results table including every skill that was tested and cut.
Stated plainly, because a skill that claims everything is useful for nothing.
name and description in the file's frontmatter, so you can also invoke it by name.A person who has closed twenty partnerships beats this file on everything except consistency. What they have that no file supplies is the read on whether a specific silence is a brush-off or a busy fortnight. What they tend not to have is a loss-reason column, because the message that fills it is the one message nobody enjoys sending.
The model with no skill is a genuine alternative here and often enough. Ask it three things: what do you send after a no, what is in the message, and what is in the reply to an invoice that is nineteen days late. If it stops selling completely, asks exactly one question, names a return date, and writes the invoice message without an apology in it, you do not need this.
A CRM covers the half this ignores. Reminders, dated next actions and a pipeline view are solved problems worth paying for, and they are orthogonal to what the message says. Buy the tool, and do not let its template library write the rejection reply, because a template is exactly the wrong register for the one message whose whole value is that it sounds like a person who has stopped trying to win.
For the money branch, a bookkeeper who chases invoices for a living is better than any of this, and cheaper than the working capital gap.
--- name: partnership-deal-desk-sequence description: Writes the partnership or business development sequence that begins at the first reply rather than at the opening pitch. Covers a three-token test for classifying any reply, including the warm one that means nothing, then the four branches almost nobody drafts in advance: the yes and the contract chase that follows it, the silence and its cadence, the refusal and the one-question message that follows it, and the unpaid invoice. Also covers arming a deputy who cannot decide, raising a rate, asking a partner to promote you, and the deal desk columns that make the branches runnable. This skill should be used when a partner, distributor, integration, sponsor or business development contact has replied and the next message has to be written, when a partnership has gone quiet or been refused, or when an agreed invoice has passed its due date. --- # Partnership deal desk sequence ## The claim this skill is built on A partnership is decided after the first reply, and the first reply is where almost every written plan runs out. Outreach preparation stops at the opening message. Everything after it is improvised, at the moment your own state is least useful for writing. A yes makes you eager, so you concede a term in the same hour you won one. Silence makes you anxious, so you send length. A refusal makes you defensive, so you argue. An unpaid invoice makes you embarrassed, so you either hedge until the due date stops meaning anything, or you skip three weeks and arrive carrying legal vocabulary. Writing the branches in advance takes your mood out of the message. That is the whole mechanism, and it is why a good opening is so often wasted: it did its job, produced a reply, and met a reply written by whoever you happened to be that afternoon. The second claim is that the branches are not four versions of one message. Each has a different goal and a different register, and a message in the wrong register fails even when every sentence in it is true. An acceptance in the register of a chase reads as doubt. **Every line of example copy below is written fresh for an invented scenario to show a shape, not to be pasted.** This teaches sequence structure, not templates. ## Part one. Classify the reply before you write a word Nothing gets drafted until the reply is classified, because the register follows from the class and the register is most of the message. **The three-token test.** Read the reply once and count how many of these it carries. - **A date.** A specific day, week, month or window attached to a next step. "Q4" counts. "Soon" does not. - **A name.** A person newly introduced, including them naming themselves as the owner of the decision. A team is not a person. - **A number.** A quantity, price, volume, term length, slot count or threshold. **Warmth is not a token and enthusiasm is not a token.** "This sounds great, we would definitely be up for finding a time" carries none of the three and commits to nothing, which is precisely why it gets filed as a yes. - **Two or three tokens: a commitment.** Run the yes branch today. - **One token: a soft yes.** Yes-branch register, ambiguous-branch mechanics. The next message has one job, to create a second token. - **Zero tokens, warm: the ambiguous reply,** the most common real case, below. - **Zero tokens, cool or deferring: treat it as silence,** not as a refusal. "Let me come back to you" is a deferral, not a decision. - **An explicit refusal: the no branch,** sent within one working day while the exchange is still live in their head. - **A reply from someone who cannot decide: the deputy branch,** below. **The you-cannot-tell branch, which is where to default.** The reply is warm, carries no token, and you want to read it as agreement. Assume you cannot tell, because the feeling that a reply was positive is not evidence that anything was agreed. The output is not the deck and not the full ask. It is a two-sentence disambiguating message: one closed question with a binary answer that forces a token into existence, plus a line giving them permission to say no. The permission line is the point, not politeness. **A clean no is worth more than a warm maybe, because only one of them has a next message.** A maybe occupies your pipeline and your forecast indefinitely and returns nothing. A no gives you the highest-value message in this file. So the disambiguating message is built to be easy to refuse. An invented illustration of the shape, for a hypothetical joint webinar: > Two questions and then I will get out of your inbox. Is something in the first week of October realistic, or is this a next-year idea? Either answer is useful, and no is a perfectly good one. **The deputy branch.** The person who replied is interested and cannot approve. Do not route around them to their manager, which costs you the only advocate you have there. Arm them instead: ask who else has to see this before it is a yes and what that person needs, then supply a forwardable block they can send onward untouched. Anything they have to write themselves is a thing that does not get sent. ## Part two. The register of each branch, on one page The ceilings are our defaults rather than findings. They exist so a branch has a shape you can check before sending. | Branch | What the message is for | Ceiling | What must never appear | | --- | --- | --- | --- | | Yes | Remove every step between agreement and signature | 120 words | A new term, a widened scope, celebration | | Ambiguous | Force one token into existence | 60 words | A re-pitch of anything already said | | Silence | Find out why, not restate the offer | 40 words, falling | A summary of the proposal | | No | One answer and one date | 80 words | Any claim about your product, at all | | Invoice | A decision, on a date | 120 words | An apology, and legal words you will not use | ## Part three. The yes branch, and the chase that follows it **Reply the same working day.** Enthusiasm decays and internal advocacy decays with it. Speed is the one term you control completely and it costs nothing. **Reinforce the decision in one sentence, in their terms.** Not thanks. Someone who has just committed on your behalf starts looking for evidence they were right, because a colleague is about to ask, and the sentence that supplies it is the one that survives the internal conversation you are not in. **Do not widen scope in the acceptance message.** Every new term reopens a decision that had just closed, and reopening costs more than the term is worth. **Move to signature in the same message and name the mechanism.** In person if you will genuinely be in the same room within days, otherwise electronic signature, and send the document rather than promising it. A document that exists gets chased. A promised one is forgotten by both sides, and neither feels late. **Start the work before the paperwork lands, and choose carefully what you build.** Finished work pulls a hesitant counterparty across a line arguments do not. Build the thing that is worthless without you: a co-branded page, a draft announcement carrying both names, a walkthrough recorded for their support team. Never build the thing with standalone value, such as the integration itself or a data export, because that is the deliverable that can be taken and used. Put a date on the draft asset, because a dated draft is a deadline the other side inherits without having agreed to one. **The contract chase: remove friction before adding pressure.** The move that beats a harder nudge is a change of mechanism. Offer to collect the signature in person where that is possible, move to electronic signature, or cut a five-page agreement to a one-page order form that references the terms rather than restating them. **When you ask, name a specific return time.** "Can you get this back to me by Friday at three?" is a commitment somebody can accept or renegotiate in four words. "At your convenience" is a task with no edge, and a task with no edge loses to every task that has one. State a time zone whenever you are unsure of theirs. **Escalate the mechanism, not the tone.** Chase one resends the signature link with a named return time. Chase two offers a short, oddly specific call, nine minutes rather than thirty, which reads as considerate and is memorable because it is not a default. Chase three offers to change the document: shorter, or a pilot with a stated end date. If none of the three moves it, the yes was never a yes. Re-classify it and run it through the no branch deliberately, because an agreement that fades produces nothing while the same deal run through the refusal branch produces a reason and a date. ## Part four. The silence branch, and our default cadence Silence with no reply at all belongs to the cold sequence that precedes this file. Silence after a reply is this branch, and it is warmer, more valuable and easier to damage. **Our default, stated as a choice and not as a finding: three follow-ups, on day 4, day 12 and day 28 after the last contact, then stop and move the counterparty to a dated re-approach roughly ninety days out.** Why that spacing. A few days reads as a nudge, a fortnight as a check-in, a month as a new conversation, so three intervals cover all three registers exactly once. Each message is shorter than the one before, because reply rate is driven by how easy a message is to answer rather than by how completely the case is made. The terminal message does not close the file, because a partner is a long-horizon asset. **Every follow-up exists to discover why they are not replying, never to restate the offer.** Restating produces nothing, because the offer was not the problem, and it turns a warm contact into someone who associates you with pressure. - **Follow-up one is the routing question.** If this has moved to somebody else, who. Someone with no interest in your proposal often answers anyway, because it costs nothing and it gets you off their desk. - **Follow-up two adds exactly one fact that is new to them.** A change on their side, a result from another partner, or the asset itself. Not a summary. - **Follow-up three changes the question from will you to when, if ever.** The highest-yield partnership form is the priority question: is this a this-quarter thing or a next-year thing. It is closed, it is easy, and its answer is a date, which is a token. - **The terminal message is one line.** You are stopping and coming back in a named month, and if that is wrong they should say so. **The rules for adjusting matter more than the numbers.** Adjust on the value of one yes: a partnership worth a quarter of revenue justifies six touches across six months, one worth an afternoon justifies two. Adjust on evidence of life: an automatic away reply, a public role change or a funding round each justify one touch beyond your default. **Any reply exits the schedule**, which is the most common automation failure in the discipline. And a visible organisational cycle, a reorganisation, a fiscal year end, their own launch, moves the whole cadence out rather than compressing it, because chasing through one spends the relationship to buy a reply nobody could have given you. ## Part five. The no branch, which is the highest-value message in the file Write this one before you need it, because you will need it in a state in which you should not be drafting. **The prohibition is absolute. Stop selling.** Not fewer claims. Zero. No line about what they might be missing, no comparison, no discount held in reserve. Continuing to pitch after a refusal tells the recipient that saying no did not work, which leaves them one move, and it is to stop reading you permanently. The quieter version of the same mistake is adding them to a mailing list without asking. Both produce a filter rule, and a filter rule outlasts the refusal by years. **Why this message works, which is the part worth understanding.** The decision has already been made. That is not an obstacle, it is the condition that makes the message function. Before a decision, everything a counterparty says is negotiation: an objection is a lever, so it gets shaded, softened, or replaced with the acceptable public reason, which is nearly always budget or timing. Afterwards none of that holds. There is nothing left for them to protect, answering costs nothing, and there is a small social debt from turning you down that one easy question is a comfortable way to discharge. So the person who would not give you the real reason while deciding will often give it afterwards, in their own words, within days. That is why it returns better material than a survey: a survey asks people with nothing at stake to recall a preference, and this asks someone who just made a real decision to name its cause while it is fresh. **The structure. Four elements, under 80 words, and nothing else.** 1. **Accept it in one sentence, with no counter-argument attached.** Any hedge, including the polite kind beginning "understood, though it is worth saying", turns a request for guidance into a rebuttal, and a rebuttal is answered with silence. 2. **Exactly one question, and it asks what would have won it.** One. Not a list, not a request for feedback in general. Two questions is a form, and a form gets postponed. 3. **A named re-approach date, our default being roughly two months out**, stated as a month rather than as "down the line". You are asking permission to come back, and a specific date is easy to decline and easier to accept than an open-ended intention. 4. **Nothing else.** No attachment, no link, no postscript. An invented illustration of the shape: > Understood, and thank you for telling me straight rather than letting it drift. One question, and there is no wrong answer: what would we have needed to have in place for this to have been a yes? I will leave it there and come back to you in early November, unless you would rather I did not. **On the response rate, stated honestly.** The operator whose method this is drawn from reported roughly two replies in five to this message, from their own sending, in one market. That is a single-account observation and not a benchmark: it was never run as a study, it has no control, and the market it came from may share nothing with yours. Take the mechanism seriously, treat the figure as a hypothesis to check against your own outbox, and keep it out of any forecast. **What to do with the answer.** Record it verbatim in a loss-reason column, never summarised. Summarised reasons converge on the same three words within a quarter and stop being worth anything. Verbatim answers stay useful: three naming the same missing capability is a roadmap item, three naming three different things is a targeting problem rather than a product one, and telling those apart is why the column exists. **The underlying diagnosis.** A refusal is usually evidence that the pitch was built around what you wanted rather than what was on their list this quarter, which points the next approach at their priorities instead of your features. **And occasionally it reverses.** Sometimes the answer names a condition you can meet this week. Take it when it happens and never aim for it, because a message written to reverse a decision reads exactly like one and stops working. The reversal is a by-product of a message that genuinely stopped selling. ## Part six. The unpaid invoice, a different conversation entirely Most people get this branch wrong for one reason: embarrassment about asking for money they have already earned. Embarrassment produces two shapes. **The apologetic hedge**, which opens with sorry to bother you and closes with whenever you get a chance, and tells the reader the due date was decorative. And **the sudden escalation**, where three weeks of silence is followed by legal vocabulary, which is the first the other side has heard that anything is wrong and turns an oversight into a dispute with positions in it. The message is not a chase. **It is a request for a decision, on a date.** 1. **Name the amount and the invoice reference in the first line.** Not "the outstanding balance". The number and the reference, plainly. 2. **Name the relationship and the work in one clause.** What was delivered and when. Not a justification, a reminder that a person is at the other end. 3. **Say what it is actually about, and mean it.** For a small supplier the line that works is usually the true one: that the money matters less than knowing where you stand. It is disarming precisely because it is not a tactic, and it stops working the moment it is one. 4. **Name a window and a consequence.** Our default window is two working days, long enough for someone who has to ask a colleague, short enough that it does not slide into the next payment run. State the consequence as a fact rather than a threat: work pauses, the next delivery is held, the account moves to your standard late-payment process, statutory interest applies where it does. **Never name a consequence you will not carry out**, because one unenforced consequence teaches the other side that all of your dates are negotiable. 5. **Close with one closed question.** Is it going out this week, or is there a problem I can help with. The second half is not softening. It is the branch where you find out about a cash flow problem early enough to arrange something. **Timing and sender.** Send it the day after the due date, not three weeks later: the first message after a date passes is cheap and unremarkable, while the identical message a month later is an accusation. If a second is needed, send it from a different role or address, because a change of sender is a change of register without any change of tone, and it signals process rather than annoyance. ## Part seven. Two adjacent asks in the same register **Raising your rate.** Open with the specific work delivered, not gratitude in the abstract. State the new rate plainly in one sentence, with no apology and no stack of justifications, because a long justification reads as a negotiating position and invites one. Set the effective date at least a month out, extend the current rate for one extra month as a stated courtesy, which turns an announcement into a concession without costing you the increase, and close with a direct question rather than an invitation to discuss. **Asking a partner to promote you.** A yes-branch message, and the ask must be smaller than the yes already given. Write the post or the email for them, supply the link, name the exact date and time, and attach a calendar invitation so the reminding is nobody's job. Send it only where a real relationship exists: sent cold it returns nothing and spends the relationship you were building. ## Part eight. The deal desk, one row per counterparty The branches are only runnable if something tracks them. One row each: counterparty and the named person who actually replied; the decision owner where that is somebody else; branch, meaning ambiguous, yes, silence, no, signed or invoice; ask size, meaning what is on the table now rather than the terminal ask; last touch date; **a next dated action, which is never blank**, because a blank next action is how a partnership dies and it dies quietly enough that nobody notices for a quarter; for a loss, the reason verbatim and the named re-approach date; and a horizon of normal or long. **The long-horizon bucket is a column, not a sentiment.** For the few counterparties who genuinely matter, budget a twelve-month relationship rather than a five-week sequence. It needs its own column for an accounting reason: a partner you are deliberately nurturing over a year is indistinguishable, in a pipeline view, from a stalled deal, and stalled deals get either abandoned or chased at a sales tempo. Both are the wrong treatment for the deals worth the most. **The loss-reason column is the asset.** It is the only column that cannot be reconstructed later, it is the entire reason the no branch exists, and it is what a pipeline review is actually for. ## Worked example, compressed A three-person team sells a time-tracking add-on. They want two things from a mid-sized invoicing platform: a listing in its integrations directory, and a co-signed email to the platform's small-business segment. The opening message has gone to the head of partnerships with the operations lead copied. **Day 2, first reply,** from the operations lead rather than the head of partnerships: interesting, they are always looking at integrations, send over whatever you have and it will get a look. **Classification.** No date, no name beyond the sender, no number. Zero tokens, warm. **Ambiguous, and also the deputy branch,** because the person who replied does not own the directory. **Day 2, disambiguating message, 51 words.** Not the deck. One closed question on whether a listing this quarter is realistic or a next-year idea, one routing question on who else has to see it, and one line saying a no is a fine answer. **Day 4, reply.** Realistically Q4. The partnerships lead owns the directory and is back from leave on the 14th. They normally need a security questionnaire completed and a named support contact. **Classification: a date, a name and a stated requirement. A commitment, but to a process rather than to the deal.** Yes-branch mechanics on the process: the questionnaire goes back before the 14th without being asked twice, the support contact is named, and a one-page draft of the co-signed email carrying both names and a proposed send date is attached. Cost absorbed, and worth nothing to the platform without the add-on. **Day 21, the yes on the listing.** Reply the same working day: one sentence reinforcing the decision in their terms, naming the support ticket topic the integration removes, the order form attached rather than promised, and a named return time of Thursday at four. **Day 35, silence on the order form.** Chase two changes the mechanism rather than the tone, offering either a nine-minute walkthrough or a one-page order form. It comes back signed on day 39. **Day 44, the refusal on the second ask.** The co-signed email is declined outright: the platform does not put partner names in its lifecycle email. **No branch, 63 words.** Acceptance with no counter-argument, exactly one question about what would have won it, one named month for the return, nothing attached. **Day 45, the answer.** It is not about the add-on. They stopped co-signing after a partner outage put their support queue under for a week. What they run instead is a quarterly partner round-up post, and applications for it open in January. **Verdict: one ask signed, the other lost with a recorded reason and a dated route back in.** The no branch produced four things nobody could have reached by guessing: the real cause, a policy formed by a previous incident rather than an opinion about this add-on; a substitute channel that was never advertised; a date in January; and the knowledge that a published uptime and support-response commitment is the specific thing that would have won it, which is a product decision rather than a marketing one. Without that message the column would read "not interested", which is not a reason. ## Failure modes **The warm maybe filed as a yes.** A reply with no date, no name and no number is recorded as agreement and forecast as pipeline. Nobody finds the error for a quarter, and the person who wrote the optimistic entry is the one who would have to correct it, which is why it stands. **Conceding in the acceptance message.** The yes arrives and the reply, out of eagerness, adds a deliverable or softens a term. A closed decision reopens, and the new negotiation starts from the concession rather than from the original position. **The courtesy chase.** The agreement goes out and the nudge asks for it back at their convenience. The request has no edge, so it loses to every request that has one, and three weeks later the sender concludes the partner went cold when nobody was ever asked for a specific day. **Selling past the no.** The refusal is answered with three more reasons and a discount. The recipient stops reading permanently, and the sender forfeits both the answer and the dated return that were available for the price of one honest message. **The quiet list add.** A refusal is followed by a newsletter subscription nobody asked for. It feels harmless and it is the fastest route to a permanent filter rule, which is a worse outcome than the refusal it followed. **The apologetic invoice.** The message opens with an apology and closes with whenever you get a chance. It communicates that the due date was optional, and the next invoice is paid later than this one was. **The three-week escalation.** Nothing is sent for a month, then the first message about a missed date arrives carrying legal vocabulary. The other side hears about the problem for the first time in its most aggressive form, and an oversight becomes a dispute. **Routing around the deputy.** The person who replied cannot approve, so the next message goes over their head. The only advocate inside that organisation now has a reason to stay quiet, and the deal has lost the one thing it had. **Nurture indistinguishable from stall.** Long-horizon relationships sit in the same view as dead deals, so they are either chased at a sales tempo, which burns them, or written off in a clean-up, which loses them. Both look like tidy pipeline management at the time. **The unsigned yes nobody re-classifies.** An agreement in principle that never becomes a document stays in the pipeline as a yes forever, because nobody wants to demote it. It produces neither revenue nor a loss reason, which makes it the most expensive row on the sheet. ## What this skill does not do - It cannot read the relationship. It classifies what a message contains, and a short reply that is a brush-off in one market is ordinary courtesy in another. The token test is a text test and it will be confidently wrong about intent. - The response rates it repeats came from one sender's own outbox in one market, with no control and no published method. Every figure here is a hypothesis to check, not a benchmark, and none of it belongs in a forecast. - It does not write or review the agreement. Exclusivity, term, IP ownership, termination and liability are lawyer work, and the yes branch drives fast towards a document this file cannot produce. - The invoice branch is one written message. What happens after the stated consequence is set by your contract and by local late-payment law, and a bookkeeper, a collections firm or a solicitor is better at that than any sequence. - It does not remember. Every branch depends on a dated next action existing somewhere outside your head, and a branch nobody is reminded to run does not exist. - It does not fix a partnership with no value on the other side. A perfectly branched sequence around an offer that does nothing for the counterparty produces a well-documented series of refusals, and the loss-reason column will tell you so if you fill it in.
These skills all ask your assistant to check things against your actual codebase, your actual schema, your actual design system. Locul keeps that context current on its own, from the files you already have, on your machine. Mac and Windows, free to start.