an agent skill · a gate before the playbook — outbound, discovery, proposals, forecasts

/sales

sales converts qualified demand into committed customers. Ask it to build a pipeline forecast and it checks your win rate before it multiplies your pipeline by a fixed number; ask it to run a discovery call, write a proposal, or design your CRM stages and it writes the real artifact, not a script; ask whether you even need an assisted sales motion yet and it answers honestly — often "not yet," naming which of four disagreeing gates would actually decide it. Account selection, cold outbound, discovery calls, demos, objections, proposals and negotiation, CRM structure, forecasting, win/loss learning, and the expansion close — 15 references and two self-testing calculators behind one router. What makes it different: it asks the question the best-known guide in this field never asks. Every stage-conversion number, win rate, and named methodology carries its source, or it doesn't ship.

# natural language — no flags, no fixed pipeline /sales our forecast says we'll close $340k this quarter — is that a real number or a guess dressed up as one?

runs onClaude CodeCodexCursorAntigravityopencodeGrok BuildHermes

The gate, then the router. There is no fixed lead-to-close pipeline to run start-to-finish — each job stands alone and enters where your request is, after the gate clears. The animation traces one path (the gate, then the forecasting-honesty check on a coverage ratio); the sections below map the whole surface it routes across.

Win the deal, honestly

Win business, and know whether you're actually winning it. sales owns the buying conversation and the deal system — account selection and qualification, cold outbound, discovery calls, demos, objection handling and enablement, proposal and pricing presentation and negotiation, CRM structure and pipeline hygiene, forecasting, win/loss learning, and the expansion-sale close — and it owns the epistemics that come with them: which forecast signals can be trusted, and which cannot. sales writes gates, playbooks, policies, pipeline structures, and forecasts. It writes no production code, and no figure it cannot source.

sales owns

everything between "will this deal close" and "what can we actually prove about that"

  • the existence gate — whether an assisted sales motion is worth running yet, and which of four disagreeing gates actually governs, before a single CRM stage exists
  • the buying conversation — account selection and qualification (ICP inherited from marketing, never redefined), cold outbound, discovery calls, demos, objection handling and enablement
  • forecasting-honesty ⭐ — whether a coverage ratio or forecast category can actually predict anything, and what a small pipeline can and cannot validate
  • the deal system — proposals, pricing presentation and negotiation, CRM structure and pipeline hygiene, win/loss learning, and the expansion-sale close

hands off to

the "Not this skill" table — six asks this skill declines by design

  • marketing — demand creation, positioning, content, deliverability doctrine, ICP definition; marketing gives up cold outbound and enablement to sales explicitly, with no hedge, but owns the qualified-demand signal and the assets sales pulls from
  • growth — funnel experiments, including on a sales-assisted stage; growth hedges where the funnel/deal line sits, never whether deal mechanics are sales's
  • success — the renewal workflow and NRR reading; sales closes the expansion sale, success runs the renewal — contested, not settled, and sales does not rule on which position is right
  • product — pricing-tier design, dunning and billing-failure recovery, pricing-page psychology; product designs the price, sales presents and negotiates a number inside a live deal
  • ai — agent mechanics, retrieval, and eval instrumentation behind an AI-SDR; sales sets the disclosure policy, ai measures compliance with it
  • automation — CRM sync and write-path mechanics; sales owns which stages and fields exist, automation owns how they stay in sync

The default reader is a founder selling their own product, not an AE carrying a quota. So the front door is a gate, not a playbook — because no source in this field settles when an assisted motion becomes worth running. Named methodologies (MEDDIC, MEDDPICC, SPIN, Challenger, Sandler) are cited and taught by their evidence, never reproduced — practitioner opinion on each is split, and this pack claims no consensus in either direction.

The existence gate — the front door

Before a single outbound sequence, CRM stage, or forecast gets written, this pack asks whether an assisted sales motion is worth running at all. No canonical source resolves that question — instead, four serious, independently arrived-at practitioner positions answer it with four incompatible gates, and the disagreement itself is the finding: a pack that adopts only one will misfire for a reader whose situation fits a different one.

gate typewhothe rule
Stage gate (count + ARR)Jason LemkinThe founder closes the first 10–20 customers personally; the transition to a sales team happens around ~$1M ARR, and the rule is hire two reps, not one, before any head of sales.
Price-tier gateChristoph JanzElephants ($100k+ ARPA), deer ($10k+), rabbits ($1k+), mice ($100+), flies ($10+) — the tier decides whether human sales is viable at all; cold calling "usually doesn't work" at the rabbit tier.
Purchasing-friction gatePatrick McKenzieThe gate is billing approval, not deal size — pricing a plan between $250 and $499/mo specifically to stay under a corporate-card or PO-approval threshold that would otherwise force a sales motion.
Founder-effort gatePaul GrahamNot about motion existence at all — founders must recruit and serve users by hand regardless of channel, whichever gate above applies.
Read the gates against each other, not in isolation: a $40/mo product with an enterprise-only compliance ask is "no sales needed" under the price-tier gate and "may still need a sales-assisted upsell path for the compliance segment" under the purchasing-friction gate. Both are right, about different axes — the pack asks which axis actually binds for the product in front of you, rather than defaulting to whichever gate is most familiar. It also names two live pressures on the gate itself: practitioners on both the buyer's and the seller's side argue the line is moving, not fixed — a build-it-yourself pressure eroding it from below, and a named condition under which a "relationship moat" on the seller's side would need revisiting, stated by the source itself rather than invented by this pack.

The ten primary jobs

Each job is one reference, read fully only when its route is selected. Forecasting honesty is the centerpiece because it is the file every coverage ratio, every board update, and every "why did we lose that deal" conversation eventually has to answer to. This is the whole buying-to-close surface, not a headline slice.

I need to…ReadContribution
Decide whether an assisted sales motion applies yet, and which gate governs when-sales-applies.md The Existence Gate. Four incompatible practitioner gates taught as disagreement, the gate eroding from below, the complexity-moat pressure with its own stated breaking condition, and the PLG usage-threshold boundary
Select or prioritize accounts, or set up a qualification framework icp-and-qualification.md ICP inherited from marketing, never redefined; frameworks (MEDDIC/MEDDPICC/BANT) vs. methods (SPIN/Challenger/Sandler) disambiguated; the current MEDDPICC trademark and litigation status
Design or run cold outbound, or set an AI-outbound policy outbound-and-prospecting.md Cold outbound owned per marketing's own unhedged handoff to sales; deliverability cited, never restated; the disclose-at-first-contact rule; the false-engagement-signal category rule
Run or structure a discovery call discovery-calls.md The rapport-evidence-vs-named-method gap: real evidence that question-asking builds liking in ordinary conversation; no published evidence any branded methodology improves win rate
Build or run a product demo demo-and-narrative.md Thin evidence stated honestly — no benchmark figure ships without a source; the demo taught as the answer to discovery, not a fixed script
Build a battlecard or objection library objections-and-enablement.md Sales enablement owned per marketing's own handoff; battlecard discipline fed by win/loss learning; no evidence any objection framework changes outcomes, stated plainly
Present pricing, write a proposal, or negotiate proposals-pricing-and-negotiation.md Pricing presentation, not design; anchor-first taught as a live disagreement with sources on both sides; discounting discipline; the expansion-sale close as a live-deal negotiation against an existing relationship
Design CRM stages or audit deal-pipeline hygiene crm-and-pipeline.md The stage/exit-criteria gap — a real process artifact the open CRM tool landscape can't express; CRM structure (sales) vs. sync mechanics (automation)
Build or audit a pipeline forecast forecasting-honesty.md Centerpiece. Assumption-disclosure is prior art, taught with credit; the four-leg no-accuracy-measurement finding; coverage as a tautology (1/win-rate); the small-numbers gate; the predict-then-score loop that works at n=1
Run or design a win/loss program win-loss-learning.md Nisbett & Wilson (1977)'s validity boundary on stated reasons; the predict-then-score corrective; no authoritative owner anywhere in the field

Full router table & invariants: SKILL.md.

Three surfaces + one additive overlay

At most one base surface reshapes every job for how the business is actually run — the same discovery-call job resolves differently for a founder taking their own calls than for a buying-committee negotiation. The agentic overlay is additive — it stacks on top of whichever base you picked, never replaces it — and carries the violet identity throughout this page, the same convention automation, operate, quality, data, marketing, growth, and success use for their own additive overlays.

Founder-led solo surface-founder-led-solo.md

The default: no reps, no dedicated CRM admin — one founder runs the whole buying conversation. If a request names no business model, this is what the skill assumes — and it says so rather than stalling for a clarification it doesn't need.

reshapeswhich jobs are worth doing by hand vs. skipping entirely · how much CRM structure is proportionate at this scale

SMB transactional surface-smb-transactional.md

A repeatable velocity motion with multiple reps — deal size and cycle are short and coached, not enterprise procurement. Qualification and forecasting get more structure than founder-led solo without the buying-committee overhead of enterprise.

reshapescoaching and playbook consistency across reps · forecast roll-up across a small team, not one person's gut feel

Enterprise/team surface-enterprise-team.md

A named CRM admin and a real buying committee — MEDDPICC-, GitLab-, and Sourcegraph-style exhibits apply directly, and buying-group reconciliation across multiple stakeholders becomes a first-class problem.

reshapesqualification-framework choice toward the frameworks built for this scale · the parallel (not sequential) sales-to-success handoff

Agentic additive

A model runs outbound, qualification, or a deal step without a human reviewing that instance — not a human approving one draft a model produced. Stacks on, does not replace. The disclosure-at-first-contact rule and the false-engagement-signal category rule don't loosen because a model runs the loop; they get more load-bearing, not less.

reshapeswhich decisions are model-decided vs. human-approved per instance · what's ai's (turn-loop mechanics) vs. sales's (the policy envelope around them)

Forecasting honesty — the centerpiece

Ask an agent to build a forecast and it either ships a single defended-sounding number, or a fixed multiple with no derivation behind it. This pack checks first: forecast method is published everywhere — a well-built forecasting pack already ships "name your assumptions or the forecast is theatre," and this pack credits it rather than claiming that idea as its own. Forecast accuracy is published by no one — not a single vendor, analyst, or academic source in this field discloses whether its own stage-weighted probability was ever checked against what actually happened.

"Forecast method is publishable. Forecast accuracy is published by no one." — a portable number, checked

What nobody discloses, at four levels

  • method ≠ accuracya public sales handbook names a detailed ten-stage process with exit criteria, but keeps its forecasting page internal and states plainly its sales metrics "are Not Public"
  • the most complete method founddiscloses stage, category, amount, and close date in full — and still carries no accuracy figure anywhere on the page
  • eight open-source CRMs, zero claimsgrepping the actual probability-computation code — not marketing copy — finds no accuracy word anywhere; most compute no probability at all, or use a static lookup table nobody validates
  • every practitioner query, silentacross this pack's research, not one practitioner stated a stage-weighted probability and said whether it was ever backtested — not defended, not attacked, simply never examined

Two peer-reviewed studies frame why: forecasts used as targets get gamed toward the target — sandbagging is documented, not folklore (Lawrence, Goodwin, O'Connor & Önkal, 2006) — and a 120-forecaster survey found only 52.8% ever check whether their own judgmental adjustments improved accuracy, with 25% using no error measure at all (Fildes & Goodwin, 2007). See forecasting-honesty.md §2–3 for the full citation trail.

A portable number, checked

"3x pipeline coverage" — the fixed multiple that circulates. No source backtests it, and it only holds at a ~33% win rate held fixed coverage = 1 / win-rate — the real identity. At a 33% win rate you need roughly 3x; at 25%, 4x; at 20%, 5x — derive it from your own numbers, never borrow the constant n = 1 forecast period — too small to calibrate on its own; a single deal's "60% probability" is not falsifiable at any volume this small
A win rate without its denominator is not a number. Coverage is derived, never shipped as a portable constant. The one calibration mechanism found on any surface in this research that works at any volume — predict the outcome before it happens, then score against what actually occurred — costs nothing and needs no CRM, no board, and no history.
a bare "70%" carries its subject inline, alwaysa forecast category (Commit/Best Case/Pipeline) is a judgment call, stated as one, not a derivation

What transfers from the best-built forecasting guidance, at no cost: naming the conversion rate, the data window, and the weighting choice behind any number — none of it requires pretending the resulting forecast was ever measured against an outcome.

What makes this different

Sales advice is not scarce — it's confident. Real playbooks cover real pieces of the buying conversation; almost none of them cite a source for the numbers they repeat, and the largest one in this field runs an unlinked stage-conversion table as executable code with its entire provenance in one unsourced sentence. What doesn't exist anywhere else is a pack that puts the whole buying-conversation surface and the honesty layer underneath it in the same repo, and grades every figure it leans on instead of repeating it because it sounds authoritative.

Real guidance, cited nowhere, integrated nowhere

The largest open sales guide found in this research (11,105 lines) cites zero permalinks to any figure it repeats. Several other packs ship win/loss content, forecasting scripts, or objection libraries by name, at single- to low-double-digit installs, none citing a primary source for the practices they teach.

  • the gate is asked first — whether an assisted motion applies at all, before the apparatus, not discovered as an afterthought once a CRM already exists
  • the honesty layer travels with the forecast — what a coverage ratio or forecast category can predict, why the small-numbers gate blocks calibration on a solo pipeline, answered with evidence rather than left as an open question
  • boundaries stated, not assumed — cold outbound vs. demand creation, CRM structure vs. sync mechanics, pricing presentation vs. pricing design — each line is named and cited, never silently claimed

Every figure carries its evidence grade, or it doesn't ship

Sales canon repeats a small set of numbers no primary source actually supports. This pack traced each one back and built a rule against shipping it again.

  • "92% give up after 4 follow-up nos" — traces to an anonymous dead blog naming no study, with circulating versions that contradict each other's own breakdowns; never cited
  • "contact within 5 minutes = 21x more likely to qualify" — the classic misattributed speed-to-lead figure, circulating with a source name and no link anywhere this research looked
  • SPIN's "17% higher volume," MEDDIC's "20–30% higher win rates" — zero-citation vendor marketing; SPIN's own 35,000-call study was never published, and no empirical study of MEDDIC was located at all
  • calculators are executable, not prose — coverage_calc.py and forecast_calibration_check.py each refuse to output a verdict without the reader's own win rate, denominator, or honest sample size
Confident, not empty. Real prior art exists — a well-built forecasting pack that already ships "name your assumptions," genuine win/loss tooling, deep enterprise qualification libraries — and is credited respectfully throughout. The gap this pack fills is that no single one of them covers the buying conversation start to finish and grades its own numbers. An honest "this has never been measured, and here's why" is a deliverable, not a failure to produce one.

The universal invariants

These govern every route, whichever references it loads.

What a pass produces

Gates, playbooks, policies, pipeline structures, forecasts, and handoffs are delivered as documents. None is delivered as production code. A sales artifact is incomplete unless it carries the gate verdict (or a statement that a sales motion already applies), every figure with its source and evidence grade, the boundary naming which sibling owns the adjacent half, and what the artifact cannot support.

step 1run the gatewhether a sales motion applies at all; then the one primary job and the base surface, overlay only if a model acts unreviewed
step 2name the decisionwhat's actually being decided, and who owns that decision once the artifact exists
step 3consume upstreammarketing's qualified-demand signal, product's pricing tiers, an ICP already defined — cited by pack and filename, never re-derived
step 4grade every figureevery benchmark or named methodology relied on, and where its evidence stops
step 5produce the artifacta gate verdict, playbook, policy, pipeline structure, forecast, or handoff — with a named proof source behind every claim
step 6run the honesty checksthe denominator check for any win rate, the small-numbers gate for any forecast, the subject check for any bare percentage
step 7hand offstate what the artifact cannot support, and emit a compact handoff when downstream work is expected

coverage_calc.py

assets/coverage_calc.py

Computes coverage as 1/win-rate and refuses output without the reader's own win rate and denominator statement — never a borrowed "3x."

forecast_calibration_check.py

assets/forecast_calibration_check.py

A predict-then-score log scorer that refuses a calibration verdict below an honest sample size — the small-numbers gate, enforced in code.

loss_prediction_log.md

assets/loss_prediction_log.md

The predict-then-score template with example rows — log a reason before the outcome is known, then check it against what actually happened.

stage_exit_criteria_audit.md

assets/stage_exit_criteria_audit.md

An audit sheet for per-stage exit criteria and negative disqualifiers — the only non-runnable asset, deliberately, since a stage list's own logic needs a human read.

Eval suites

evals/ — 46 cases

27 activation + 10 traversal + 6 output + 3 compression-ablation cases: does the router reach the smallest sufficient reference set, and does a never-ship figure ever resurface, even inside a correction of it.

The handoff boundaries

sales operates independently when invoked alone, and uses compatible upstream artifacts without silently overriding them. handoff.md maps every boundary between this pack and the rest of the family from sales's side — including three rows siblings pre-drew for sales before this pack shipped, mirrored back exactly rather than re-hedged.

sales writes

gate_verdict: <applies /
  lighter-weight / not-yet>
job_and_surface: <which one,
  and any surface assumed>
claims: [claim, named proof source,
  evidence tier]
boundary: <sibling pack + filename
  owning the adjacent half>
cannot_support: <what the
  artifact does not prove>

marketing

Pipeline and close-rate signal that feeds positioning and message iteration. Sales consumes qualified demand — a defined fit-plus-intent signal on a named account — and the assets sales pulls from. Sales does not create demand or own the ICP definition; it inherits marketing's ICP-as-a-funnel model, stated exactly as marketing itself states it.

growth (provisional)

Pipeline and qualification signal shaping which stage of a sales-assisted funnel is worth testing. Sales consumes an experiment readout on a funnel stage; cold outbound, CRM stages, and deal mechanics stay sales's regardless of which funnel stage an experiment touches.

success (contested)

Sales hands success the expansion sale, closed in a sales-assisted motion. Sales consumes the renewal workflow up to a complex deal needing negotiation. Not settled: success's own position holds a CSM renews by default and only genuinely complex renewals route out — sales does not rule on which position is right, and states both sides intact.

A worked example, not a ruling. One public enterprise handbook shows a CSM introduced during the technical-evaluation stage and full ownership transferring at closed-won, with formality scaling by deal size — but its own text states the sales and success roles run parallel, not sequential, and the company is one enterprise, high-ACV instance. It corroborates the contested split above; it does not settle it. product and automation each already give up related work to sales in their own text, without a matching row here yet — a gap this pack notes rather than fills on their behalf.

The rest of the family — each an independently installable pack with its own guide:

Start here

Install once. It's a plain SKILL.md router — no flags, no config, no fixed pipeline — so it activates on natural-language phrasing ("do we need an assisted sales motion yet," "audit our forecast before we present it," "draft our AI-outbound disclosure policy") rather than a fixed command.

# skills.sh ecosystem npx skills add gabros20/sales-skill -g -y # clone + manual copy git clone https://github.com/gabros20/sales-skill cp -R sales-skill/skills/sales ~/.claude/skills/sales # use — natural language, any host /sales do we need an assisted sales motion yet, or is self-serve conversion covering it fine /sales audit our pipeline forecast before we present a coverage ratio to the board /sales draft our AI-outbound disclosure policy — what has to be disclosed, and when

The same install runs on any Agent Skills host. Codex triggers with $sales; manual copy into any client's skills directory also works.

what's in the repo
skills/sales/ the skill: SKILL.md (router) + 15 references/ + 4 self-testing assets/ + evals/ research/ multi-channel research + build-gate synthesis site/ this guide — deploys to salesskill.vercel.app README.md · SOURCES.md · LICENSE

More detail: SKILL.md · SOURCES.md — source attribution, licensing rule, and the numbers this pack refuses to ship.