Function-first strategy design and reuse
Use for product strategy discussions, existing-funnel reconstruction, and experiments assembled from an existing funnel of screen variants. Keep a discussion-only request read-only. Local guidance maintenance does not authorize strategy writes, subtype registration, publication, or paid research runs.
Read the functional model first
| Function | User-facing job |
|---|---|
| F1 | Confirm that this is the right place and continue the acquisition promise. |
| F2 | Learn the user's goal or context so the path can become relevant. |
| F3 | Obtain information necessary to deliver the service or configure the purchase. |
| F4 | Guide the user to value through ordering, routing, skips, and transitions. |
| F5 | Build motivation by showing credible relevance to the user's needs. |
| F6 | Let the user preview or try on the product's value before committing. |
| F7 | Present a relevant offer and handle readiness, purchase, or refusal. |
F4 is a property of the flow. Research is also cross-cutting: answers and exits can inform hypotheses without requiring a dedicated research screen. Functions are not a mandatory sequence of seven blocks. A block is a conversation topic; screens implement it. One screen may realize several functions, and the same function may appear in several blocks.
Set strategy-wide context before selecting blocks
Separate two decisions in the strategy brief:
- Delivery environment: a web funnel or onboarding inside an installed app. This sets shared interaction, loading, navigation, and platform constraints.
- Entry context and intent: for example, interrupted social browsing, search, a returning visitor, an email link, or first launch after installation. This sets the acquisition promise and the working assumptions about readiness, familiarity, patience, and the effort already invested.
Web does not imply social traffic; installed app does not imply purchase intent. For AURA the combined profile is web plus a social UGC ad entry, with uncertain intent and an easy return to the feed. Record that profile once on the strategy. Pass its applicable rules into every block brief, subtype selection, screen brief, and review. Blocks inherit shared rules and specify only their local application or justified exceptions, rather than duplicating the global profile.
Keep the functional model common across environments. A reusable block/subtype declares where it applies and may provide web and in-app implementations. Split subtypes when responsibility or the interaction contract changes, not merely to repeat the strategy's platform flag on every library entry.
The CLI already carries strategyType (including Web Onboarding and
IOS Onboarding); set the supported value explicitly rather than accepting the
app-oriented default. The two decisions above are authoring concepts, not a
claim that a new structured entry-profile API or automatic rule inheritance
already exists. Verify runtime propagation before relying on it, and keep
schema, prompt, and UI changes as separate implementation work.
Apply the social-to-web pacing profile
Distinguish a web funnel entered from social browsing from onboarding after an app installation. For a social-ad entry, the click may express curiosity rather than purchase intent; the person can return to the feed immediately. Installing an app represents an additional deliberate action, but does not prove readiness to buy either. Set the entry assumptions from the actual source and user behavior: not every web visitor is cold and not every app user is committed.
For the AURA social-to-web case, treat attention as fragile and earn each next interaction through a visible payoff. This is a cross-block design constraint:
- Continue the creative's promise, person, product, and visual story on arrival. Render useful content immediately; a matching first-frame poster can hold the composition while video loads. Do not gate the first action on media playback.
- Prefer a simple tap and an immediate visible response over a detached brand introduction, multi-screen tutorial, or questionnaire before value appears.
- Combine compatible functions when it lowers effort: welcome plus color preview can establish relevance, collect an order choice, and enable a first try-on.
- Make every answer visibly affect a relevant next step. Keep required purchase data and useful personalization distinct; defer high-effort asks until justified.
- Keep a quick rhythm through low latency, clear choices, concise content, and meaningful feedback. Avoid forced reading timers, fake processing, or animation that delays interaction. Preserve user control and readable pacing.
- Let the story demonstrate use and resolve relevant objections progressively. Do not import every possible reassurance beat from a longer app onboarding.
- In the approved sales flow, keep the offer legible and payment effort low; give ready users a clear next action while preserving voluntary refusal.
When reusing a system example, review its assumed entry commitment, effort, waiting, and time until visible value. Keep the function; adapt the interaction to the web context instead of copying an app sequence mechanically. The strategy owns the shared entry assumptions; the block/subtype records its compatibility and local consequences. Establish timing targets through observation and experiments; do not invent a universal seconds-per-screen rule or promise a conversion lift.
Check the generated funnel against the entry profile and pacing constraints before relying on it.
Design the conversation before assigning screens
Start from the actual product, entry promise, available facts and assets, required order/service data, and existing screen behavior. Describe the block's purpose and observable outcome before choosing its type, subtype, or screen count.
A welcome should establish relevance, continue the story, and start a useful conversation. Choose a simple question the person can answer immediately. Its answer should configure something necessary, make later steps more relevant, or return a clear immediate benefit. Do not ask for a context variable with no identified consumer, or classify optional segmentation as hard-required data.
An opener need not be a separate greeting followed by a goals question. A product choice can combine F1, F3, and F6 when it preserves the entry promise and shows the chosen item immediately. If the user explicitly chooses this approach, record it as a hypothesis, preserve confirmed values, and skip duplicate asks. Do not infer needs from a color choice. If context-based personalization still needs a use-case answer, collect it separately or specify an honest fallback.
For a physical product, distinguish use context (usually F2) from a selected sellable variant (F3 when the order needs that choice). A color-matched product story can combine F5 and F6. It should also address relevant fears, blockers, and objections using supported product facts and demonstrations, selected for the declared use context. Mark unresearched objections as hypotheses; do not invent capabilities or proof to answer them. The product's actual order contract decides what is required; do not make color universally mandatory for every product.
If the agreed sales responsibility is one paywall block, keep its main offer, checkout, recovery/downsell, upsell, and paid/unpaid completion as internal screens or branches. They do not each become a strategic block merely because they have different screen IDs. Keep success, cancellation, failure, expiry, and refusal explicit; a screen or simulated animation is not payment evidence.
Give each agreed block a reusable contract
Record only what applies to the block:
- purpose, declared functions, and visible user benefit;
- entry state and acquisition context;
- required and optional inputs, outputs, and their downstream consumers;
- invariant behavior and allowed implementation alternatives;
- internal branches, boundary exits, fallbacks, and completion conditions;
- evidence versus operator assumptions, and the hypothesis being tested;
- source funnel/version/screen provenance and exact approved assets;
- observable success criteria and the proposed experiment's changed variable.
Separate binding behavior from recommended presentation. For example, confirmed color propagating to every product image is an invariant; Portrait, Swatchbook, and Palette are presentation alternatives. A change to the collected data or payment path requires a contract review, even when the visual change is small.
Reuse the user's library and preserve subtype identity
Distinguish a visual library funnel, a project subtype, a native block example, and a block instance in a strategy. They can refer to one another but are different records. Do not require a native-library migration simply to discuss or reuse a visual library.
When the user requests a library entry and subtype per block, define a stable semantic subtype for each agreed reusable block, with its purpose, playbook, binding rules, and examples. Keep visual skins and ordinary layout alternatives as implementations of that subtype unless their responsibility or contract changes. Do not use a style name such as Cinema as the sole semantic definition.
The project custom-subtype catalog is exposed by
segmently strategies subtypes list <projectId> and
segmently strategies subtypes apply <subtypeId> --file <file> <projectId>.
Inspect installed command help and current records before use. The current
catalog is for custom blocks; do not invent a subtype field on a built-in block.
Verify supported binding fields before saving and read back both the subtype
and its block/example association. A display label or tag is not that binding.
Prepare the concrete catalog change for review; do not assume this command has
the native example library's checksum/dry-run flags.
Use screen-block-builder for native example capture/validation/write/readback.
Preserve original full custom screens as exact source artifacts. Native
previewScreens are render-only; adaptation uses the simplified blueprint and
asset intents. Do not claim that adaptation or insert-as-is will preserve opaque
HTML, JavaScript, purchase callbacks, or existing media exactly. For exact reuse,
verify the supported full-screen clone/lift/update route and its readback on one
block before applying it across the library. Retain source provenance without
misassigning active ownership to the source strategy.
Work without audiences or actors when that is the actual case
Two valid authoring paths have different evidence:
- Hypothesis-led: use the product brief, source creative, existing screens, explicit operator decisions, and declared variables. Author the strategy manually through the supported strategy/block/screen commands without fake insight pins. Mark unresearched needs and effects as hypotheses. Do not invent actors, interviews, prevalence, testimonials, or insight citations.
- Research-led: define tasks and candidate audiences/actors, run the supported research process, review the resulting insights and variables, then pin the selected analysis for generation. Generated actors and simulated interviews remain model-based hypotheses, not observed customer behavior or causal proof.
Research for a virtual product is a useful optional experiment; lack of prior research does not invalidate manual strategy work. The Product CLI currently accepts existing actor selections for insights generation; it does not expose audience, actor, or interview generation. Inspect the supported product workflow before promising an automated end-to-end research run. Discussing this option does not authorize starting it or changing the shared project's audience mode.
The current automatic generation path is insight-first and its initial prompt prescribes Welcome plus User Goals. Manual metadata creation without pins is supported. These are not interchangeable capabilities. A merged welcome/color opener or no-insight automatic generation needs explicit runtime verification; never fabricate inputs or silently run the default pipeline instead. The older motivation guidance allows only an F2 data-collection opener: record an explicitly requested F1+F3+F6 opener as a scoped hypothesis, not an unnoticed global rule change. Updating this skill does not update or deploy those runtime prompts.
Discuss and experiment in stages
For a step-by-step request: agree on the block map, then the first block's contract and subtype, then its reusable implementations. Advance only to the stage the user has authorized. Save/register the reviewed records when requested; keep strategy creation and traffic publication separate.
An experiment may change block ordering, combine functions into fewer screens, replace a block implementation, or change screen content. Name its level and preserve all other agreed contracts. Pin the baseline and the hypothesis before building the alternative. Library previews and test purchases verify behavior; they do not establish business lift.