Turn one leaking conversion flow — a landing page, a signup, a checkout, an in-app onboarding — into a\nmeasured, tested funnel. This is an interactive, resumable journey of seven phases: the agent asks\nbefore every decision and records the outcome in your project's docs/ folder, so you can stop after\nany phase and pick up later. It works on websites and inside products alike; the unit of work is the\nflow and its ONE action, not the whole site.
Find the leak with numbers, learn the reason from customers, fix message → offer → proof → friction\nin that order, and prove every fix with a pre-committed test. The order is causal: a number tells\nyou where, only research tells you why, motivation must be raised before friction-cutting pays,\nand an untested fix is a guess that compounds. This skill sequences the phases, asks the decision\nquestions, and records every choice in docs/. The constituent skills carry the method — invoke them\nrather than improvising their frameworks.
| Phase | \nSkill | \nQuestion it answers | \nArtifact | \n
|---|---|---|---|
| 1 | \nlean-analytics | \nWhere does the flow actually leak, and what is the one metric? | \nExtends docs/METRICS.md — GATE | \n
| 2 | \ncro-methodology | \nWhy do people drop at the leak — which objections and friction? | \nCreates docs/FUNNEL.md; extends docs/EXPERIMENTS.md — GATE | \n
| 3 | \nstorybrand-messaging | \nDoes the leaking step promise the visitor's own desired outcome in five seconds? | \nExtends docs/POSITIONING.md + docs/FUNNEL.md + docs/EXPERIMENTS.md | \n
| 4 | \nhundred-million-offers | \nIs the offer at the conversion point worth acting on now? | \nExtends docs/OFFER.md + docs/EXPERIMENTS.md | \n
| 5 | \ninfluence-psychology | \nIs there honest proof at every point of doubt? | \nExtends docs/FUNNEL.md + docs/EXPERIMENTS.md | \n
| 6 | \ndesign-everyday-things | \nCan a visitor who decided to act complete the flow without stumbling? | \nExtends docs/FUNNEL.md + docs/DESIGN.md + docs/EXPERIMENTS.md | \n
| 7 | \ncro-methodology | \nWill we know the fix worked — pre-committed metric, sample size, no peeking? | \nExtends docs/EXPERIMENTS.md + docs/METRICS.md | \n
docs/CONVERSION-OPTIMIZATION-PLAN.md and every artifact in the Journey Map. If the tracker exists, summarize the journey state in 3-5 lines and ask which phase to enter. Done when the user has confirmed an entry point. A journey with a tracker is resumed, never restarted.docs/CONVERSION-OPTIMIZATION-PLAN.md with every phase statused pending | in-progress | awaiting-evidence | done | deferred: reason | skipped: reason. Done when the tracker exists and the user has confirmed the phase plan.in-progress on proceed. Done when the user chose.npx skills add wondelai/skills/<slug> --global. If the user declines, run the phase from its Brief — the minimum viable method. State which mode you are in.done.docs/. Every recommendation lands as a checkbox or a table row with owner and priority. See references/artifact-templates.md when creating a docs/ file for the first time — create it from the full skeleton (all section headings), then fill the sections your phase names.Ask these before creating the tracker:
\nSkip heuristics: skip Phase 3 when messaging was already validated (e.g. an improve-website journey completed its message phases); skip Phase 4 when the offer is fixed by contract — record skipped: reason. Phase 7 may be deferred: reason at very low traffic in favor of before/after evidence with an explicit revert trigger, never silently skipped. Never skip Phases 1-2 — an unfound leak and an unresearched reason turn every later phase into guessing.
Then create docs/CONVERSION-OPTIMIZATION-PLAN.md from the template and confirm the plan. Done when the tracker exists with every phase statused and the user has confirmed the plan.
Phases run in the listed order — each assumes the previous phase's artifact exists. Any phase can be entered, skipped, or deferred per the Operating Rules, but Phases 1-2 gate them all: nothing downstream fixes a problem that isn't a measured leak with a researched reason. When running any phase from its Brief (constituent skill not installed), read references/methods.md first — it carries each phase's full method, checklists, formulas, and benchmarks; the Brief is only the summary.
\nPurpose: Locate where the flow actually loses people and pick the one metric this journey moves — before any opinion about why.
\nBrief (fallback): A good metric is a comparative ratio that changes what you do next; totals and\ncumulative charts are vanity. Express each step of the flow as a conversion rate, compare against your\nown history and published benchmarks (e-commerce converts ~1-3% of visitors; landing pages on paid\ntraffic low single digits), and find the biggest absolute drop on the highest-value path. Pick the One\nMetric That Matters for this journey, pair it with a counter-metric so it can't be gamed (signup rate\n× 30-day retention), and draw a line in the sand: target, date, pre-committed miss response. Cohort\nand segment (channel, device, plan) — one collapsing segment hides inside a flat average.
\nInvoke: Use the lean-analytics skill with the flow steps and analytics from intake. Ask for a step-by-step funnel table with baselines and benchmarks, the OMTM plus counter-metric for this journey, and the biggest leak ranked by absolute lost value.
Decide with the user: (1) Confirm the OMTM and its counter-metric. (2) Which leak to attack first — biggest absolute loss on the money path, not the easiest percentage. (3) If instrumentation is missing, which minimal events to add first — the phase stays awaiting-evidence until the numbers exist.
Artifact: Extend docs/METRICS.md ## Funnel (stage | conversion | benchmark | bottleneck?), ## Stage & One Metric That Matters, and ## Baselines & Targets. Update the tracker.
Done when: the funnel is measured at the coarsest granularity that still localizes the leak to a single step, the OMTM and counter-metric are recorded with a line in the sand, and the leak is named — only then are Phases 2-7 unlocked. Finer sub-steps awaiting instrumentation stay awaiting-evidence in Next Actions and do not block the journey, provided the named leak does not depend on them.
Purpose: Replace guesses about the leak with evidence from real visitors. Phases 3-6 may only fix problems traceable to a finding here.
\nBrief (fallback): Don't guess — discover. Mine primary sources (a one-question exit survey:\n\"What's preventing you from [action] today?\"; post-conversion: \"What almost stopped you?\"; chat logs,\ntickets, sales calls) and secondary sources (reviews, competitors) for the customer's own words. Sort\nobjections into the Big 5 — Trust, Price, Fit, Timing, Effort — and build the O/CO table: every\nobjection gets an evidence-backed counter placed at the exact step the doubt arises, never in an FAQ.\nDiagnose each step with the LIFT lenses (value proposition ± clarity, relevance, urgency, minus\nanxiety and distraction — Goward) and the MECLABS heuristic (conversion rises with motivation and\nvalue clarity, falls with friction and anxiety). Rank fix hypotheses by ICE and apply the 10x screen:\nif a change couldn't plausibly double the step, don't queue it.
\nInvoke: Use the cro-methodology skill with the Phase 1 leak, the flow, and the voice-of-customer sources from intake. Ask for the researched objection list in customer words, the O/CO table with placements, missing persuasion assets, and an ICE-ranked hypothesis backlog.
Decide with the user: (1) Which researched objection is the primary leak driver. (2) Low traffic: accept qualitative plus heuristic evidence — explicitly. (3) Which implicit objections need CO-Only counters (countered without being stated).
\nArtifact: Create docs/FUNNEL.md with ## Flow Map & ONE Action, ## Leak Diagnosis, and ## Objections & Counters (O/CO); extend docs/EXPERIMENTS.md ## Experiment Backlog (ICE-ranked). METRICS.md ## Funnel stays canonical for the conversion numbers — Leak Diagnosis cites them and adds the researched reason and severity. Update the tracker.
Done when: the flow map names the ONE action per step, every researched objection has an evidence-backed counter and a placement, and the backlog is ICE-ranked.
\nPurpose: Make the leaking step say what the visitor gets, in their words, in five seconds — clarity converts before persuasion can.
\nBrief (fallback): The customer is the hero; you are the guide. Run SB7 on the leaking step: a\nCharacter who wants one thing, their Problem at three levels (external, internal — the frustration the\ncopy must name, philosophical), you as the Guide (empathy + authority), a 3-step Plan that makes\nacting feel safe, one Direct plus one Transitional CTA, and named failure/success stakes. Rewrite the\nstep's headline in customer language pulled straight from Phase 2 — customer words outperform\ncopywriter words. Then make it stick (Made to Stick): concrete beats abstract (\"save 16 hours a\nmonth,\" not \"boost productivity\"), and pick the single Commander's Intent message the visitor must\nstill remember tomorrow.
\nInvoke: Use the storybrand-messaging skill with the step's current copy, the Phase 2 objection evidence, and POSITIONING.md if it exists. Ask for above-the-fold rewrites that name the internal problem, a one-liner, and one Direct plus one Transitional CTA.
Decide with the user: Which rewrite passes the 5-second test (a stranger can say what's offered and why it matters); which internal problem the copy names; whether the step keeps a transitional CTA or goes single-CTA.
\nArtifact: Extend docs/POSITIONING.md ## Brand Script (StoryBrand), ## One-Liner, and ## Key Messages (surface | message | status); record the rewrite in the Proposed message/CTA column of docs/FUNNEL.md ## Flow Map & ONE Action, leaving Current message/CTA intact as the before-state Phase 7 measures against; append copy tests to docs/EXPERIMENTS.md ## Experiment Backlog. Update the tracker.
Done when: the leaking step has a rewritten message that names the internal problem, one primary CTA, and a logged test hypothesis.
\nPurpose: Strengthen what is actually exchanged at the conversion point — the best flow cannot sell a weak offer, and the offer is the biggest single lever.
\nBrief (fallback): Value = (Dream Outcome × Perceived Likelihood) ÷ (Time Delay × Effort &\nSacrifice). Raise the numerator with outcome language and proof; crush the denominator with speed\n(\"first result in 5 minutes\") and done-for-you framing. Reverse the risk with a guarantee aimed at the\ntop Phase 2 objection — it raises perceived likelihood and lowers anxiety at once. Stack named,\nhonestly-valued bonuses that each kill one objection; present price after value, anchored against the\nstack. Scarcity and urgency only when real — fake deadlines convert once and churn forever.
\nInvoke: Use the hundred-million-offers skill with the current offer from intake and the Phase 2 Price/Timing/Effort objections. Ask for a Value Equation score per lever, a guarantee design matched to the top objection, and a trim-and-stack pass on the offer components.
Decide with the user: Which guarantee the business can actually honor; which bonuses are real and sustainable; whether price presentation changes (anchoring, payment plans) — pricing itself may be out of scope; record that.
\nArtifact: Extend docs/OFFER.md ## Offer Stack (element | description | honest value | objection it kills), ## Price Metric when the flow raises what the price is charged per, plus ## Willingness-to-Pay Evidence if new evidence surfaced; append offer tests to docs/EXPERIMENTS.md ## Experiment Backlog. Update the tracker.
Done when: each Value Equation lever has a concrete change or a reason it stays, the guarantee targets the top researched objection, and every offer change carries a test hypothesis.
\nPurpose: Answer each remaining objection with honest evidence placed at the exact step the doubt arises.
\nBrief (fallback): Under uncertainty people use shortcuts: social proof (specific numbers — \"2,347\nteams\" — and similar-others beat generic praise), authority (credentials, certifications), commitment\n(a small first yes makes the big yes consistent — ask for the card after the value moment, not\nbefore), reciprocity (give the useful thing first), and scarcity (real only). Audit the flow for proof\ngaps: every O/CO row needs its counter actually rendered — testimonial, data point, logo bar,\nguarantee seal — at its placement. Proof hierarchy: specific results with context > named testimonials\nwith faces > case studies > statistics > logos. The transparency test gates everything: if knowing the\ntechnique would make the visitor feel tricked, it fails. Fabricated proof and hidden costs are\ndefects, not tactics.
\nInvoke: Use the influence-psychology skill with the FUNNEL.md O/CO table and the flow's current proof. Ask for a per-step proof audit, which principle answers each open objection, and copy for the two highest-impact placements.
Decide with the user: Which proof assets exist versus must be acquired (testimonial requests, case studies — log acquisition as Next Actions); where scarcity or urgency claims are genuinely true; which steps get a commitment micro-ask.
\nArtifact: Extend docs/FUNNEL.md ## Proof Inventory (asset | type | placement | status) and complete the ## Objections & Counters (O/CO) placements; append proof tests to docs/EXPERIMENTS.md ## Experiment Backlog. Update the tracker.
Done when: every open objection row has a real proof asset placed or an acquisition task with an owner, and no claim in the flow is unverifiable.
\nPurpose: Protect the visitors who decided to act. B=MAP: behavior happens when Motivation, Ability, and a Prompt converge — Phases 3-5 raised motivation; this phase raises ability and sharpens prompts.
\nBrief (fallback): Bridge Norman's two gulfs. Execution: clear signifiers (buttons look pressable,\nfields look editable) and constraints that make errors impossible (date picker over free text, submit\ndisabled until valid). Evaluation: feedback within 0.1s of every action, progress indication on\nmulti-step flows. Forms (Baymard): every field costs conversions — cut to the minimum, ask for payment\nas late as possible, show all costs before the final step (surprise shipping and taxes is the #1\ncheckout killer), offer guest checkout, validate inline with messages that say how to fix. Error\nmessages state what went wrong and how to fix it, without blame; slips get undo, not are-you-sure\ndialogs.
\nInvoke: Use the design-everyday-things skill with the conversion-critical steps (form, payment, confirmation). Ask for weak signifiers, where constraints replace error messages, feedback gaps, field-by-field form cuts, and message rewrites.
Decide with the user: Which fields are truly required now versus collectable later; where a constraint replaces a warning; whether the flow shows total cost earlier; which severity-4 friction items ship immediately as a logged before/after card versus wait for a full test.
\nArtifact: Extend docs/FUNNEL.md ## Flow Friction Audit (step | issue | severity 0-4 | fix | status); extend docs/DESIGN.md ## UX Audit Findings for reusable component fixes, naming the Norman gulf (execution or evaluation) in the Heuristic column; append fixes to docs/EXPERIMENTS.md ## Experiment Backlog. Update the tracker.
Done when: every step has a friction audit row, forms are cut to minimum fields, severity-4 items have owners, and error messages meet the checklist. Where money changes hands in this flow, all costs are visible before the final step and guest checkout is offered; in a flow that takes no payment, mark those two rows n/a.
Purpose: Turn the high-ICE fixes into trustworthy experiments — without rigor you cannot tell a real lift from noise, and a false winner compounds forever.
\nBrief (fallback): Pre-commit everything (Kohavi): primary metric, guardrail metrics (the Phase 1\ncounter-metric, revenue per visitor, support volume), sample size computed from baseline rate and\nminimum detectable effect at 95% significance / 80% power, and a duration of at least one full\nbusiness cycle covering weekdays and weekends. Never peek and stop early — it manufactures false\npositives; never rerun until you like the answer. Test bold changes — meek tweaks rarely reach\nsignificance. Low traffic: run a before/after window with qualitative confirmation, or ship as a\nreversible bet with an explicit revert trigger — and say which you are doing. Practical significance\ngates rollout: a statistically significant 0.1% lift may not pay for its complexity. Winners become\nthe new control; losers become learnings written down.
\nInvoke: Use the cro-methodology skill with the top ICE hypotheses from the backlog. Ask for full experiment designs: hypothesis in \"If [change], then [metric] because [research]\" form, pre-committed sample size and duration, and the decision rule.
Decide with the user: Which 1-3 hypotheses run first (highest ICE on the money path); the guardrail metrics; what happens on a flat result — iterate the fix, or return to Phase 2 for better research.
\nArtifact: Promote backlog rows to docs/EXPERIMENTS.md ## Experiment Cards, adding the pre-committed sample size per arm, planned duration, and the minimum lift worth keeping as extra bullets on the card (the skeleton's bullets carry hypothesis, metrics, and decision rule; extra bullets are additive and allowed); record results and verdicts as they land; extend docs/METRICS.md ## Baselines & Targets with a new dated row per tested metric, leaving the pre-test baseline row intact. Update the tracker.
Done when: each shipped fix is a card with a pre-committed metric, sample size, and decision rule; results carry verdicts; and the OMTM's new baseline is written down.
\n| Skill | \nAdd when | \nArtifact | \n
|---|---|---|
| ux-heuristics | \nthe whole page confuses, not just the flow — visitors stumble before reaching the conversion steps | \nExtends docs/DESIGN.md ## UX Audit Findings | \n
| microinteractions | \nthe flow's moments feel dead — silent taps, unexplained waits, abrupt state changes | \nExtends docs/DESIGN.md ## Microinteraction Inventory | \n
| scorecard-marketing | \nthe flow needs a lower-commitment entry — a quiz or assessment as the transitional conversion | \nExtends docs/WEBSITE.md ## Lead Capture | \n
| hooked-ux | \nthe conversion sticks but the next visit doesn't — post-conversion activation is the real leak | \nExtends docs/PRODUCT.md ## Hook Model | \n
Optional phases follow the same operating rules — load and use each listed skill exactly as a core phase would; insert where the Add-when condition first becomes true. They carry no inline Brief: standalone, run ux-heuristics as a severity-rated pass over Nielsen's 10 heuristics, microinteractions as a trigger/rules/feedback/loops inventory, scorecard-marketing as a quiz-funnel design ending in a personalized result, and hooked-ux as a trigger → action → variable reward → investment loop audit — or install the named skill for its full framework.
\n| Mistake | \nFix | \n
|---|---|
| Redesigning the step before measuring where the leak is | \nRun Phase 1 first — the biggest drop is rarely where opinion points; optimize the money path, not the loudest complaint. | \n
| Guessing objections instead of mining customer words | \nExit surveys, tickets, and sales calls (cro-methodology) — teams are almost always wrong about why visitors leave. | \n
| Polishing persuasion on top of an unclear message | \nClarity before psychology: pass the 5-second test (storybrand-messaging) before adding proof and urgency. | \n
| Treating the offer as fixed and testing only cosmetics | \nThe offer is the biggest lever (hundred-million-offers); a guarantee change outlifts a button change by orders of magnitude. | \n
| Faking scarcity or cherry-picking proof | \nConverts once, churns forever, caps trust permanently — real scarcity and verifiable proof only (influence-psychology). | \n
| Adding form fields \"while we're at it\" | \nEvery field costs conversions (Baymard); collect later what you don't need now (design-everyday-things). | \n
| Peeking at test results and stopping early | \nPre-commit sample size and duration, then report whatever comes back (cro-methodology, Kohavi). | \n
A funnel always has a next-biggest leak: when the Phase 7 verdicts land, re-enter Phase 1 with fresh numbers rather than declaring victory — the journey is a loop with an exit condition, and the exit condition is the line in the sand from Phase 1, not exhaustion.
\nExit checklist — every box tied to an artifact:
\ndocs/CONVERSION-OPTIMIZATION-PLAN.md is done, deferred: reason, or skipped: reason.Close the tracker: remaining Next Actions carried into FUNNEL.md and EXPERIMENTS.md so nothing is lost. Then route forward: when the flow converts and needs more qualified traffic, continue with the grow-website skill; when the leak has moved past conversion into engagement and retention, continue with the improve-app skill; when the whole site — look, typography, speed, message — needs the broader pass, continue with the improve-website skill.