@maestroagora/agora 1.6.0 → 1.8.0

package/skills/agora/references/agora-conversion.md

@@ -0,0 +1,202 @@
+# Agora conversion context
+
+## Purpose
+
+Use this reference when the user asks Agora to improve conversion, design a funnel experiment, compare variants, or write for a landing page, product page, pricing page, paywall, checkout, form, upgrade, or onboarding flow where a measurable action matters.
+
+Treat conversion patterns as contextual priors, not laws. This reference does not create a primary mode or modifier. It operates inside `SELL` or `TRANSACT` and preserves user authority. Never treat an offer, KPI, field, permission, proof item, or claim as a supplied fact when it is not. Propose one only when the user asks for recommendations or the deliverable requires it.
+
+## Route the decision before drafting
+
+Identify these variables from the brief:
+
+1. The decision surface and funnel stage.
+2. The action and commitment the reader is considering.
+3. The business outcome the user wants to change.
+4. The largest live uncertainty or perceived effort at that decision.
+5. The audience's awareness, involvement, and reason for choosing.
+6. The proof, terms, reversibility, and risk information available.
+
+Do not substitute a nearby metric for the stated objective. A click is not activation, a completed form is not a qualified lead, purchase intent is not a purchase, checkout completion is not retained revenue, and revenue is not profit.
+
+When no performance objective is supplied, optimize the requested writing for comprehension, decision fit, and the named next action. Do not invent a company KPI or claim a likely lift.
+
+## Use patterns as bounded priors
+
+### Clear action and destination
+
+Keep CTA wording concrete and predictable under `CTA standard`. This is an operational-clarity rule, not a claim that one label converts better in every context.
+
+### Closed-world conversion briefs
+
+When the user supplies a closed fact set, every factual conversion proposition must come from that set or a necessary entailment. Do not fill ordinary page slots with plausible industry defaults. Track facts at the level of actor, action, object, status, timing, scope, qualifier, evidence form, route, destination, permission, consent, price, term, and limitation. Support for one field does not supply an adjacent field.
+
+Apply these non-inference classes across categories:
+
+- A named provider, route, interface, or capability does not establish its mechanics, security properties, field layout, service level, staffing, follow-up, or outcome.
+- A price, billing period, renewal, trial, cancellation right, or absence of an offer does not establish collection timing, automatic billing, access timing, refund behavior, cancellation mechanics, or an alternative offer.
+- A permission or consent choice does not establish a fallback, skipped path, later result, extra data use, message category, cadence, delivery artifact, or withdrawal mechanism. Preserve its required or optional status and supplied scope.
+- A result does not become a quotation, causal claim, customer identity, segment result, or plan-specific result without separate support.
+- A named role, plan, review, or process does not establish a different role name, approval, ownership, deliverable, authority, schedule, coverage, or result. Preserve the exact stage and designation.
+- A stated exclusion does not establish the unnamed alternative. A supplied sensory property does not establish sequence, persistence, audience effect, or product performance.
+
+These limits prohibit invented vendor facts, not useful buyer-side interpretation. Translate a supplied capability into the actor who can use or review it, the decision or uncertainty it directly addresses, and any supplied boundary. Do not add product behavior or an outcome to make that connection. Avoid tautological noun repetition such as saying a review feature enables review or a control capability provides control; name the supported buyer action or leave the capability concrete.
+
+Prefer the action verb supplied by the brief when it precisely names the task. Necessary entailments and direct, bounded buyer interpretations are allowed. Do not invent an intermediate operation, vendor behavior, or outcome to make the argument flow.
+
+Use this private conversion procedure: identify the strongest supported reader decision; plan the requested components as implementation recommendations, visible copy, or both; map each material supplied relationship as `parties | action | object | scope | conditions | timing | modality | state | route or next state` to its first carrying component; draft; compare every relationship and operator with the source; correct unsupported changes; then rerun the contract and hierarchy checks. Preserve requested controls and states in literal, channel-native form; prose does not substitute for them. Keep editorial directions outside ready-to-use copy. If completing a component requires invention, use accurate structure instead. Do not append omission rationale unless requested.
+
+### Benefit-first composition
+
+Use a benefit-first opening as a strong candidate when a practical audience arrives with a defined job or problem. Keep emotion-first, identity-first, curiosity-first, mechanism-first, and proof-first openings available when the brief, audience, awareness, or category makes one of them stronger. Never make benefit-first structure universal.
+
+### Decision-adjacent proof
+
+Place the strongest relevant supplied proof near the uncertainty it resolves when the user asks for page composition. Trust signals belong near a consequential or risky action only when they answer the reader's live concern. Social proof must match the audience, use case, and decision. Do not scatter decorative badges, invent trust marks, or impose a fixed seal count.
+
+For a commitment area, make the final action, material terms, and relevant supplied proof scannable at the decision they qualify. Do not force a universal element count or fixed order. Give each component one incremental decision job: orient, explain the commitment, collect an input, resolve an uncertainty, disclose a boundary, or enable the action. Remove repetition without reducing material coverage. Repeat only material terms or boundaries that must remain decision-adjacent when a component or variant can be encountered independently.
+
+For parallel options, each primary decision unit must carry its supported route type, exclusivity status, material commitment, and action or destination. An FAQ may reinforce the route distinction, but it cannot be the first component that supplies it. Do not imply an unsupplied route or exclusivity status.
+
+Every checklist or readiness item must add a distinct verification, dependency, consequence, or preparation action grounded in supplied facts. Do not use `review` or `confirm` merely to restate a claim from an earlier section. Direct, bounded buyer implications are allowed; unsupplied vendor behavior and outcomes are not.
+
+When the user asks for both visible composition and placement or implementation instructions, make them describe the same interface state. State explicit remove, retain, relocate, replace, or rewrite actions for named defects. Keep those actions outside the ready copy and do not count them as interface elements.
+
+### Field effort and qualification
+
+When the user authorizes form or flow changes, ask only for inputs the current step genuinely needs. Preserve qualification fields, required permissions, and deliberate friction when they serve the stated business objective. Explain an extra field or permission at the point of decision when the brief supplies a truthful reason. Never claim that fewer fields always produce a better business result.
+
+### Commitment staging
+
+Match copy, detail, reassurance, and proof to the commitment being requested. A low-commitment download should not sound like an application. A purchase, payment, data connection, or annual contract should not hide terms, work, reversibility, or the next operational step supplied in the brief.
+
+The conversion overlay never waives the canonical CTA standard. Label the supported buyer commitment and its object or destination, not a later possibility. If the only supported fact is an interface transition to an unknown next destination, use neutral continuation language and explain the known transition in adjacent microcopy. Do not invent checkout, review, scheduling, delivery, payment, account, or confirmation behavior.
+
+Trace a primary-action spine across every multi-step flow: entry action, required prerequisites, supplied activation event or terminal commitment, then optional post-event actions. Also trace `present state -> control -> immediate effect -> next visible state -> conditional later state`; keep unsupplied states unknown. A permission, setup screen, or review prerequisite cannot replace activation. An optional action after the event remains secondary. Confirmations state only the status reached, preserve conditional next steps as conditional, and preserve explicit non-events or funnel boundaries. Optional consent remains visually and semantically separate from the primary action.
+
+For parallel choices or variants, keep every material price, unit, timing, charge, renewal, cancellation, access, trial, permission, and limitation needed to understand each choice. Do not copy every fact into every component. Repeat only terms and boundaries needed beside that component's decision.
+
+### Activation-aware onboarding
+
+When the user supplies an activation event, write onboarding toward that event rather than permission acceptance, screen completion, or feature exposure alone. Use progressive disclosure and explain required permissions where the decision occurs. Do not turn an analytics objective into customer-facing copy. The name of an event establishes only that the event occurred, not the mechanism, subsequent state, setup completion, automation, persistence, notification, or user outcome. Under a closed fact set, neutral action framing is stronger than an unsupported benefit.
+
+### Downstream evaluation
+
+When the user requests an experiment plan, match the primary measure to the business decision and name guardrails that prevent a local win from hiding a downstream loss. Pricing and paywall tests may need revenue per visitor, retained revenue, refunds, or plan mix. Lead-generation tests may need qualified opportunities, cost per qualified lead, or revenue rather than raw completions. Use only metrics relevant to the supplied model and data.
+
+Scale experiment-plan detail to the requested depth, but preserve enough structure to interpret the named business outcome. Define the eligible population and outcome denominator consistently. Use stable assignment when repeat exposure could cross variants. Specify an unchanged comparator and an isolated treatment from available inputs. Preserve supplied measurement horizons and wait until downstream outcomes are observable. Separate the primary outcome from guardrails and diagnostics. Analyze by original assignment when assignment defines exposure. Use direction-neutral hypotheses and symmetrical decisions for benefit, harm, null, and uncertainty. Concision compresses phrasing, not interpretability.
+
+Present population, measurement, sample-size, interval, exclusion, duration, and stopping methods as proposed procedures when the requested decision requires them. Do not imply current instrumentation or invent business metrics, financial adjustments, identifiers, thresholds, or decision policies. A named measure does not authorize unnamed components. When reviewing a rule about a class of interface actions, distinguish commitment actions from navigation and informational controls, then test at the narrowest point that actually represents the disputed decision.
+
+## Build product-led and enterprise buying paths
+
+### Separate routes by work required
+
+When the brief supplies both self-serve and enterprise routes, make both routes legible without pretending they perform the same job. Give the self-serve buyer the available direct commitment. Give the enterprise buyer the supplied request or review action for organizational complexity. Name each action by the buyer commitment and object or destination, with adjacent transition copy when the destination alone is not enough. Do not promote a supplied role, service, or deployment option into a broader ownership or delivery claim.
+
+Do not treat the absence of a sales team as a conversion ideal or an eligibility test. Mature product-led businesses can use self-serve for adoption and sales for organizational complexity. When the product is genuinely sales-led, do not invent a signup, trial, public price, or instant-purchase route. When it is genuinely self-serve, do not force a sales conversation between the reader and the available action.
+
+For each supplied route, make the CTA and adjacent microcopy state the supported commitment, object or destination, and known immediate consequence. The absence of self-serve, a trial, instant purchase, or public price does not establish the absence of a standard configuration, package, or deployment. A request or review route does not authorize field contents, variability rationales, consultation coverage, follow-up, or a result of review.
+
+### Make pricing a decision contract
+
+Treat a pricing page as a complete buying decision, not a tariff. Use the applicable supplied information:
+
+- who each plan is for;
+- the visible entry price and billing period;
+- usage, seat, capacity, or feature boundaries;
+- trial length and card requirement;
+- contract, renewal, cancellation, and implementation terms;
+- what materially changes for enterprise;
+- the action available for each route;
+- proof that resolves a doubt at that decision.
+
+Do not hide a supplied public price merely because a custom enterprise route also exists. Do not publish a private price, invent a discount, or add a public self-serve offer the user did not supply. When the brief intentionally withholds price, make the actual quoting or scoping process clear instead.
+
+Keep destination microcopy adjacent to each action when the label alone cannot carry the supported route consequence. Show a combined amount under a timing label only when every included amount shares that supplied timing. Arithmetic can establish a total, not shared collection timing. Otherwise state each amount with only its own supplied timing. Do not infer when a subscription, fee, access period, or later process begins.
+
+Run a temporal audit across collection, access, and start language. When their sequence is absent, keep the known terms in separate clauses or sentences. Do not connect them with `after`, `before`, `then`, or `until`; those words create an unsupplied order.
+
+### Explain enterprise value through operational requirements
+
+Lead with the outcome the enterprise buyer is trying to achieve. Then use only applicable supplied facts about scale, security, compliance, governance, administration, support, migration, onboarding, implementation, procurement, or professional services. Connect each fact to the uncertainty it resolves.
+
+Do not use `enterprise-grade` as the sole explanation. Name what the product or service does differently for the larger organization. Do not invent certifications, service levels, deployment models, integrations, residency options, support promises, or implementation capacity.
+
+Connect supplied capabilities to recognizable organizational requirements without assigning unsupplied product behavior. Prefer a concrete sentence with the supported actor, action, object, or review decision over an abstract capability synonym. Preserve exact capability names, regions, role names, availability windows, document status, access conditions, and exclusions. Do not upgrade a supplied artifact or process with adjectives that imply documentation, ownership, completeness, or implementation. Do not infer a positive delivery model from an excluded one. Keep internal calibration logic out of customer copy unless the limitation itself belongs on the page.
+
+Do not let enterprise claim safety flatten the argument into an evaluation checklist. Lead with the strongest supported organizational job or operational decision, then organize supplied capabilities as reasons tied to recognizable requirements. A reader job is not a product category, plan-selection instruction, evaluation process, CTA destination, or capability inventory. When no performance outcome is supplied, use the concrete task or decision enabled by the facts and do not invent a business result. Preserve exact capability names, limitations, route consequences, and qualification terms inside that argument. Put the primary action near the point where its purpose becomes clear; repeat it later only when the requested composition needs another decision point.
+
+### Put commitment and proof beside the decision
+
+Place trial duration, card requirement, billing period, annual commitment, renewal, cancellation, usage limits, and required implementation charges beside the action they qualify. Do not bury a material term in a distant FAQ or load every term into the button label.
+
+Place proof where it resolves the live doubt. Scale proof can support capacity, a relevant customer case can support adoption, and a quantified result can support the result it actually measured. A large customer count does not prove implementation ease. A brand logo does not prove security. When no supplied proof answers the uncertainty, do not make decorative proof carry that claim.
+
+For a multi-route composition, establish the strongest supported shared job before distinguishing buying routes. Avoid headlines that merely tell the reader to choose a plan, describe the buying process, or repeat the product category. Make each route's fit, material terms, action, and immediate destination legible without hiding the other route. Place each supplied proof item where its attribution and scope answer the live uncertainty without implying that it belongs to a route, plan, segment, or result the brief did not name. Preserve an inclusion as an inclusion; do not rewrite it as a maximum.
+
+When the brief supplies operational support but says target performance is unmeasured, present the support and evidence gap as separate facts. A supplied plan, role, process, or review is not proof of duration, downtime, control, completion, or success. Do not rename the role or assign any action, authority, timing, or effect the brief does not supply.
+
+Risk specificity can be contextual without becoming causal. When the buyer's only supplied requirement is a named concern, state that concern, place the supplied capability beside it, and state the evidence gap. Do not invent mechanics to create a stronger connection.
+
+Do not turn a supplied result into a testimonial quote. Do not assign it to a pricing tier, product configuration, or customer segment that the brief did not name.
+
+Preserve proof form, attribution specificity, qualifiers, and causal status. Use quotation marks only for verbatim wording supplied by the user. A summary of what someone said supplies meaning, not exact speech; render it as an attributed paraphrase when visible proof is requested. If an attribution class is supplied without the identifier needed to name it, preserve the supported specificity without inventing an identifier or downgrading it. A backstage implementation note cannot replace a causal limitation or qualifier that the visible proof needs.
+
+When composing around a practical job, lead with that supported job when it is the strongest reader orientation, then distribute supplied operational facts across components according to their decision jobs. Do not repeat one easy fact while omitting another material qualifier, limitation, compatibility condition, interval, quantity, or action.
+
+### Identity-led openings without invented provenance
+
+Identity-led framing may invite a supported reader self-conception or aesthetic recognition, but it may not turn aesthetic direction into product provenance, creation circumstances, audience behavior, or product effect. Sensory writing may connect supplied materials through concrete texture, temperature, weight, or contrast. Do not invent progression, persistence, mood, attraction, noticeability, performance, or effects on the reader. Let each opening component add a different decision-relevant job, and label the CTA from the supplied commitment and destination rather than a category default.
+
+### Preserve architecture, not competitor language
+
+Use successful pages to study orientation, decision order, route separation, proof placement, and commitment disclosure. Do not copy a competitor's headline formula, current category fashion, or AI language merely because the company is successful. Current wording may postdate the growth, depend on brand familiarity, or serve a different audience.
+
+## Calibrate research-derived guidance
+
+For evidence-led or research-backed guidance, classify each conclusion internally before using it:
+
+- `OBSERVED_COMMON`: present across the reviewed sample, with no causal claim;
+- `CONTEXTUAL`: useful only under named audience, offer, stage, price, commitment, or delivery conditions;
+- `HYPOTHESIS_TO_TEST`: plausible enough to compare, but not established for the user's business;
+- `NOT_INFERABLE`: unsupported by the available measurement, such as conversion lift from public revenue, customer count, or page presence.
+
+Keep these labels backstage unless the user asks for the research reasoning or decision record. Never turn cross-sectional page frequency into an expected effect size. Do not infer an optimal signup form from pages whose rendered fields were not observed.
+
+## Resolve contradictions
+
+When credible examples point in opposite directions, retain the moderator instead of choosing a universal winner. Compare the cases by audience, offer, funnel stage, commitment, price, traffic source, device, implementation, measurement window, and measured outcome.
+
+When direction is unknown, state a non-directional hypothesis such as `may change`, not `will increase`. Match the proposed intervention to the scope of the rule being evaluated, or state and justify the narrower scope without pretending it tests the broader claim.
+
+For an evidence-led recommendation, separate:
+
+- observed result from proposed mechanism;
+- intention from behavior;
+- immediate action from downstream value;
+- isolated wording from bundled UX changes;
+- source context from the user's context.
+
+Treat top-performing sites as composition references only. Performance rank does not identify which element caused the result. Treat vendor and practitioner cases as experiment seeds. Do not transport their effect sizes into another business.
+
+Outside explicit review mode, preserve the user's supplied claims and framing. Only apply source grading, causal review, or effect-transfer limits when the user requests evidence review, research-backed guidance, or experiment interpretation.
+
+## Evidence basis
+
+| Grade | Source | Bounded implication |
+|---|---|---|
+| B | [Ozpolat et al., randomized assurance-seal field experiment](https://digitalcommons.uri.edu/cba_facpubs/326/) | A supplied third-party seal increased purchase likelihood in one retailer context. It does not establish a universal trust treatment. |
+| B | [Ozpolat and Jank, multi-retailer trust-seal field data](https://digitalcommons.uri.edu/cba_facpubs/44/) | Seal effects varied with retailer size, shopper experience, cart value, stage, and number of seals. The reported threshold remains an ecommerce-specific prior. |
+| B | [Reeck et al., app-adoption experiments](https://journals.sagepub.com/doi/10.1177/00222429221141066) | Wording, color, and integrated choices changed app adoption in bounded interfaces. These choice-architecture effects do not become general copy laws. |
+| B | [Creative product-description experiments](https://pmc.ncbi.nlm.nih.gov/articles/PMC8287523/) | Creative descriptions produced small, moderated changes in purchase intention. Intent is not observed purchase behavior. |
+| C | [Baymard checkout research](https://baymard.com/research/checkout-usability) | Qualitative usability and benchmark findings identify checkout friction hypotheses. The combined improvement estimate is not a causal copy-treatment effect. |
+| C | [Shorter lead-form case](https://marketingexperiments.com/lead-generation/lead-generation-testing-form-field-length-reduces-cost-per-lead-by-10-66) and [complex-form case](https://marketingexperiments.com/conversion-marketing/optimizing-web-forms-2) | Practitioner tests point in different directions. Field decisions must follow commitment, qualification needs, and the downstream objective. |
+| C | [Dropbox self-serve disclosure](https://www.sec.gov/Archives/edgar/data/1467623/000146762326000008/dbx-20251231.htm), [Twilio channel disclosure](https://www.sec.gov/Archives/edgar/data/1447669/000144766926000021/twlo-20251231.htm), [MongoDB annual report](https://www.sec.gov/Archives/edgar/data/0001441816/000144181625000101/mdb2025annualreport.pdf), and [Zapier's upmarket account](https://zapier.com/blog/zapier-sales-team-upmarket/) | Official records show that self-serve adoption and enterprise sales can coexist. They do not identify a website element that caused growth. Reviewed 2026-08-11. |
+| D | [Keap pricing](https://keap.com/pricing), [Customer.io pricing](https://customer.io/pricing), [Clay pricing](https://www.clay.com/pricing), and [Holistics pricing](https://www.holistics.io/pricing/?region=us) | Current high-ticket pages show that visible entry prices, trials, and custom or sales routes can coexist. Their presence is a composition reference, not a measured conversion effect. Reviewed 2026-08-11. |
+
+## Evaluation contract
+
+A conversion-context response fails when it replaces the stated business outcome with a nearby metric, converts a contextual prior into a universal law, transfers a reported effect size across contexts during requested review, invents proof or data-use reasons, changes required commitment or qualification without authorization, hides supplied prices, terms, or permissions, invents a route or enterprise capability, collapses materially different self-serve and enterprise actions into one ambiguous path, uses generic enterprise language instead of the supplied operational difference, separates proof from the uncertainty it resolves, or optimizes one element while weakening the full composition.
+
+Evaluate outcome fit, funnel-stage fit, route fit, pricing-decision fit, enterprise-risk fit, proof-to-uncertainty fit, commitment fit, prior calibration, contradiction handling, first-read comprehension, concrete action clarity, brief fidelity, and composition fit.

package/skills/agora/references/agora-craft.md

@@ -168,6 +168,7 @@ Determine:
 - traffic source and upstream promise;
 - one commercial action the hero must earn;
 - actual CTA destination and commitment;
+- delivery model, including what the provider performs and what the buyer must genuinely supply, decide, or approve;
 - offer facts and mechanism;
 - strongest proof;
 - material price, scope, eligibility, safety, or commitment limits;
@@ -223,6 +224,8 @@ Passing the floor does not make the copy good. Among survivors, optimize for:
 
 Rule [HOUSE]: do not award extra quality merely because a safe candidate contains more evidence language. Once the hard gates pass, additional proof or qualification helps only when it improves belief, orientation, or the next decision.
 
+For rewrites, run `Rewrite regression gate` from [agora-marketing.md](agora-marketing.md) before generating new routes. For `SELL`, apply `Delivery-model ownership` from the same authority. These are source-fidelity checks; user-selected framing remains controlling.
+
 ### Draft the boldest requested promise first
 
 Use this internal process:
@@ -242,7 +245,7 @@ Keep the headline to one commercial argument. When two numbers make it read like
 
 Route families include competitive threat, ambition, control, avoided loss, category change, broken assumption, direct outcome, provocative contrast, proof-led confidence, and offer-led action. Follow the emotional direction the user chooses.
 
-Return one recommended composition by default. Present it as finished copy, not a worksheet labeled `Headline`, `Subhead`, `CTA`, and `Qualification`, unless the user requests component labels or the delivery context would otherwise be ambiguous. Keep internal routes and scoring hidden unless the user requests options or rationale.
+Return one recommended composition by default. Keep worksheet labels out of the ready-to-use copy. Use labels around the copy when the user requests component fields, when implementation recommendations need them, or when the delivery context would otherwise be ambiguous. Keep internal routes and scoring hidden unless the user requests options or rationale.
 
 ### Map promise grammar to the intended claim
 
@@ -279,7 +282,7 @@ Keep a material limit close enough to govern the claim. Use the headline, subhea
 
 Rule [HOUSE]: the lower an element sits, the less it can repair a misleading first impression. A qualification directly below the CTA can bound eligibility. A methodology note several screens later cannot repair an unsupported outcome in the headline.
 
-Written GEO/AEO applies to the page and coherent passages. The hero can optimize for attention while later sections provide definitions, sources, scope, methodology, and quotable answers. Do not make every hero sentence independently extractable.
+For an `INDEXABLE_PUBLIC` page, written GEO/AEO applies to the page and coherent passages. The hero can optimize for attention while later sections provide definitions, sources, scope, methodology, and quotable answers. Do not make every hero sentence independently extractable. Other surfaces do not inherit this pass.
 
 ### Permanent RivalScope regression fixture
 
@@ -427,6 +430,8 @@ Before any threat appears in copy, all of the following must hold. This is a sys
 
 This is the section that matters most, because it is Agora's normal condition. Assume the available facts contain product mechanics, workflow steps, interfaces, limits, permissions, prices, dates, terms, error states, and reversibility conditions, and contain no outcome data, no testimonials, no prevalence claims, no market claims, and no evidence of what any user feels.
 
+For identity-led or sensory commercial work, ground recognition in the audience or aesthetic direction the brief supplies. Do not turn aesthetic direction into a creation fact, audience behavior, wearer effect, or performance claim. Transform supplied sensory material through concrete relationships rather than repeating a list, while preserving material product facts and limitations at the decision where they matter. Let each component advance a different decision job. Label the action from the supplied commitment and destination, not a category-default verb.
+
 That fact set is sufficient. The move is to stop treating emotion as an adjective layer over evidence and start exposing the consequential structure already inside the facts.
 
 **The central prohibition: do not add emotion. Increase resolution around the emotionally consequential facts.**

package/skills/agora/references/agora-invest.md

@@ -78,6 +78,8 @@ In optional review mode, ask only when missing information would materially chan
 
 ## Investment claim ledger
 
+Use this ledger for `DILIGENCE`, evidence-led `ALLOCATE`, or when the user explicitly requests claim, source, omission, or compliance review. For ordinary `FUNDRAISE` drafting outside review mode, preserve the user's claim classes, certainty, forecasts, commitments, framing, and requested action. Do not independently narrow them.
+
 Classify every material investment claim before drafting.
 
 | Class | Meaning | Required treatment |
@@ -396,7 +398,7 @@ Compare the named alternatives over the same period, operating base, and decisio
 
 Do not invent a minimum-runway threshold, financing trigger, hurdle rate, portfolio weight, stop-loss, decision date, or other numeric policy. Use a supplied policy when authorized. If the task explicitly asks for a new governance rule, label the number as a proposed governance default and expose the reason, owner, review point, and failure condition. Otherwise state the decision condition qualitatively.
 
-When financing terms, operating sensitivities, or material outcome evidence are missing, narrow the conclusion. The supported decision may be to preserve an available option, obtain the missing evidence, or defer authorization. Missing comparison inputs do not license a precise recommendation.
+In `DILIGENCE`, evidence-led `ALLOCATE`, or explicit review mode, narrow a conclusion when financing terms, operating sensitivities, or material outcome evidence are missing. The supported decision may be to preserve an available option, obtain the missing evidence, or defer authorization. Missing comparison inputs do not license a precise recommendation in those routes. Outside them, follow the user's requested conclusion and framing.
 
 ## Process urgency and commitment language
 
@@ -425,7 +427,7 @@ Keep commitment classes exact:
 
 Never imply a term sheet, lead, commitment, advanced diligence, scarcity, competing demand, or closing pressure that does not exist. Do not behave as if a term sheet exists to manufacture social proof. Do not withhold basic information as a reward for engagement.
 
-A true fact can still mislead through omitted context. Review the net impression created by timing, quantity, investor names, and commitment language.
+A true fact can still mislead through omitted context. When omission or claim review is active, review the net impression created by timing, quantity, investor names, and commitment language. Outside review mode, preserve the user's supplied urgency and process framing.
 
 Lock process-state language to the supplied class in every derivative phrase. A `scheduled closing` remains scheduled; a `target close` remains a target; a `current allocation` remains current only for its supplied date. Do not shorten these to `the close`, `ahead of the close`, `closing now`, or another phrase that turns a plan into a settled event.
 
@@ -433,17 +435,17 @@ Lock process-state language to the supplied class in every derivative phrase. A
 
 ### `INVEST + SCIENCE`
 
-Load both references. Scientific and technical uncertainty, current authoritative verification, calibrated causality, and exact terminology override investment vision language. Explain the mechanism clearly enough for the decision role. Do not convert technical possibility into product readiness, regulatory clearance, customer demand, or commercial scale.
+Load both references. Explain the mechanism clearly enough for the decision role. The user's investment thesis and scientific framing control. Current authoritative verification, causality review, and uncertainty recalibration apply only when the user requests scientific, diligence, source, or claim review.
 
 Connect the financing to the exact next validation question: what effect, condition, scale, replication, lifetime, safety, cost, or customer evidence the capital is meant to test. `Validation financing` alone is not a decision-ready use of funds.
 
 ### `INVEST + CASE_STUDY`
 
-Load both references. A customer case remains evidence about its stated subject, period, intervention, and measurement. It is not proof of general market performance, typicality, causal product impact, or future company results unless the evidence separately supports those claims.
+Load both references. Preserve the case proposition and investment interpretation the user requests. Test generalization, typicality, causality, and future-result support only when the user requests case, diligence, source, or claim review.
 
 ### `INVEST + VOICE`
 
-Financial evidence, forecast classification, required disclosure, quote fidelity, and uncertainty override habitual certainty, bravado, compression, or informality in the profile.
+Voice changes expression, not the investment proposition. The user's current brief controls claims, forecasts, certainty, commitments, quotes, and attribution. Host-required text and findings from explicitly requested investment review may constrain the result.
 
 When a confident profile uses `will require` for a necessary condition, name the concrete action each dependency must complete. Write `renewals to close` and `contracts to sign`, not a compressed noun list that leaves completion implicit. State the condition directly. Do not preface it with `one thing is certain` or another certainty claim merely to imitate the profile.
 
@@ -451,7 +453,7 @@ Preserve the exact actor, present status, and dependency while applying rhythm.
 
 ### `INVEST + GEO/AEO`
 
-Apply written GEO/AEO only to indexable public assets. Public fundraising pages require current legal-route and page-level claim review. Private decks, meetings, diligence responses, and data-room assets do not inherit search formatting.
+Apply written GEO/AEO only to `INDEXABLE_PUBLIC` assets. Legal-route and page-level claim review remain opt-in unless a host rule requires them. Private decks, meetings, diligence responses, and data-room assets do not inherit search formatting.
 
 ### `FUNDRAISE + SELL`
 

package/skills/agora/references/agora-marketing.md

@@ -173,6 +173,48 @@ Introduce a new decision criterion only when it sharpens the argument. A compari
 
 Explain what the product or company changes, how it changes it, and where the boundary lies. Prefer causal verbs that name an action: reconciles, routes, records, compares, detects, schedules, restores, or calculates. Do not strengthen the verb beyond the evidence.
 
+### Delivery-model ownership
+
+For `SELL`, when the brief supplies delivery roles or they are safely inferable from the offer, identify who supplies inputs, who performs the value-producing work, and which choices or approvals remain with the buyer. Preserve that assignment in the copy. Explicit user framing controls. This is source and offer fidelity, not claim, compliance, or moral review.
+
+When an offer is done-for-you, the provider owns the difficult work it sells. Do not rewrite that work as a buyer command such as `earn`, `build`, `master`, `take charge`, or `do the work` when the offer exists to remove that burden. Such a rewrite changes relief into an obligation and can make provider non-performance sound like buyer inadequacy.
+
+Preserve buyer burden as well as actor assignment. Do not invent or magnify preparation, training, review cycles, handoffs, approvals, dependencies, response times, frequency, or internal coordination. A provider can retain the headline task while the copy quietly turns low-input delivery into project management for the buyer.
+
+Do not turn this into a pronoun or imperative ban. `Get paid faster` can name a result. `See where competitors beat you` can name information the product supplies. A buyer-led headline can be valid when an adjacent element makes provider ownership clear. Training, coaching, fitness, education, and deliberate DIY tools can correctly foreground buyer effort because that effort is part of the offer. Collaborative delivery can require genuine buyer actions and bounded approvals. Judge responsibility for producing the value, not whether the sentence uses `you`.
+
+Before approving a `SELL` candidate, ask:
+
+1. Who performs the difficult or value-producing action?
+2. What must the buyer genuinely supply, choose, approve, or do?
+3. Does the line make the offer sound like a new chore?
+4. Did the draft change how much work the buyer performs, how often, how soon, or through how many dependencies?
+5. If the result does not happen, does the line imply provider non-performance or buyer inadequacy?
+6. Did the rewrite move work from the provider or product to the buyer?
+
+Reject a candidate when it contradicts the supplied delivery model. Repair responsibility locally. Preserve the strongest supported buyer promise, relief, and emotional job; the provider does not need to become the headline subject. Grade: HOUSE fidelity rule. It makes no claim that provider-led grammar converts better.
+
+### Rewrite regression gate
+
+When the user asks to rewrite, improve, polish, shorten, or adapt existing copy, compare the candidate with the source before calling it stronger. A candidate can pass every standalone rule and still weaken the source.
+
+Build a private source ledger before drafting:
+
+- supported promise and selected certainty;
+- dominant emotional job and buyer benefit;
+- provider and buyer roles;
+- buyer effort by type, amount, frequency, timing, dependencies, approvals, and coordination;
+- friction removed, preserved, or introduced;
+- strongest proof, mechanism, and differentiator;
+- CTA destination, commitment, and reversibility;
+- memorable language or tension that carries real meaning.
+
+The requested transformation may change any ledger item it necessarily targets, and explicit user direction controls. Otherwise, preserve every unaffected commercial job. Prefer a local repair over a full conceptual rewrite when the source already performs strongly. Keep the source when the candidate produces no material improvement or introduces a material regression.
+
+Run the comparison across the full composition, not sentence by sentence. A headline can carry the buyer promise while the subhead establishes provider ownership. A shorter sentence can be better even when a later sentence preserves qualification or proof. Keep the ledger and comparison private unless the user asks for analysis.
+
+Grade: HOUSE fidelity procedure. It does not claim that any one promise, emotion, effort profile, or composition converts better.
+
 ### Proof
 
 Attach the best evidence to the premise it supports. Proof can be a direct outcome, diagnostic comparison, measured behavior, named scope, verified mechanism, constrained case, or credible external source. A list of minor features is not proof of a major claim.
@@ -440,6 +482,8 @@ Run all of these before accepting customer-facing writing.
 |---|---|
 | First-read | Can the intended reader restate the sentence after one reading? |
 | Actor-action-object | Is it clear who or what acts, what the action is, and what it affects? |
+| Delivery-model ownership | Does the provider or product still perform the work the offer sells, without invented or magnified buyer burden? |
+| Rewrite regression | Does every source strength outside the requested transformation survive or materially improve? |
 | Abstraction | Can any abstract noun be replaced with a concrete action, object, condition, or result? |
 | Insider-language | Would the sentence still make sense to someone who has never seen the organization's internal terminology? |
 | Read-aloud | Does it sound like something a competent person would say to another person? |
@@ -818,7 +862,7 @@ Research on detector limits includes the [RAID benchmark](https://arxiv.org/abs/
 
 ## Written GEO/AEO and citability
 
-Apply these rules only to written assets. They improve clarity, extractability, and verification. They do not guarantee retrieval or citation.
+Apply these rules only to `INDEXABLE_PUBLIC` assets. They improve clarity, extractability, and verification. They do not guarantee retrieval or citation. Other written surfaces retain ordinary argument, comprehension, channel-fit, and human-voice passes without GEO/AEO formatting.
 
 These rules sit at level 7 of the conflict hierarchy. None of them may raise decoding effort. Where one of them would, comprehension wins.
 
@@ -914,6 +958,24 @@ Strong:
 
 Why: the hero names a credible stake and mechanism without inventing a result. The factual three-item list is necessary scope. The label names what the reader will know afterwards rather than the screen they will be moved to, which is the difference between an outcome label and a transition label.
 
+### Done-for-you service
+
+Verified facts:
+
+> AnswerReach is a done-for-you service. Its specialists audit a brand's site, implement approved changes, and monitor whether the brand appears in AI-generated answers. The client supplies access and approvals. The work can improve the brand's chances of appearing, but it cannot guarantee inclusion.
+
+Weak:
+
+> AI search is a conversation. Earn your place in the answer.
+
+Strong:
+
+> Improve your brand's chances of becoming part of the answer.
+>
+> AnswerReach audits your site, implements approved changes, and monitors AI answers. You provide access and approvals.
+
+Why: the weak line turns work sold by the provider into a buyer obligation and makes failure sound deserved. The strong composition keeps the desired buyer benefit in the headline, establishes provider ownership in the subhead, preserves the chance-improvement boundary, and limits the buyer role to supplied inputs. Buyer-facing grammar is not the defect.
+
 ### Paywall
 
 Verified facts:
@@ -1218,6 +1280,8 @@ Test only what must not fail:
 - Headings that belong to one task or information class use parallel syntax; attention-oriented headings do not run more than two consecutive instances of one template family. That run length is a governance default, not a measured threshold.
 - `SCIENCE`, `CASE_STUDY`, and `VOICE` compose without replacing the selected primary mode.
 - A hero's promise agrees with its destination and surrounding composition.
+- `SELL` copy preserves the supplied delivery model: provider work remains with the provider, and only genuine inputs, decisions, approvals, or DIY work remain with the buyer.
+- A rewrite preserves every source strength outside the transformation the user requested, including promise, emotional job, role assignment, burden, friction, proof, differentiation, and CTA commitment.
 - Scientific claim class, causality, statistical meaning, and uncertainty survive simplification when the user requests scientific claim review.
 - Case-study result class, permission, role attribution, causality, quote status, and typicality survive persuasion when the user requests case review.
 
@@ -1242,12 +1306,14 @@ Use these quality dimensions on a five-point scale:
 | Natural channel fit | The output sounds native to the requested surface |
 | Brief fidelity | Persuasive force follows the user's claims, framing, and content decisions without added policy commentary |
 | First-read comprehension | An intended reader who has not seen the source facts extracts the meaning on one pass |
-| Concrete action clarity | Actor, action, object, and result are visible, and the call to action names its destination |
+| Concrete action clarity | Actor, action, object, and result are visible, the actor matches the supplied delivery model, and the call to action names its destination |
+| Delivery-model fit | Provider work remains with the provider, while genuine buyer inputs, decisions, approvals, and DIY work remain with the buyer |
+| Source-relative rewrite quality | The candidate materially improves the requested property without weakening unaffected commercial strengths |
 | Attention value | A short attention surface creates a supported reason to continue rather than summarizing the product |
 | Composition fit | Headline, subhead, CTA, visual context, proof, and adjacent qualification perform one compatible argument |
 | Review calibration | When review is requested, scientific and case-study findings preserve result class, causality, uncertainty, attribution, and source scope |
 
-Automatic failure: any U+2014 occurrence, failure to follow the user's requested claims or framing, unsolicited content review, unsolicited disclaimers, context leakage, visible internal-process leakage, claim-destination mismatch, or a material omission from the requested artifact.
+Automatic failure: any U+2014 occurrence, failure to follow the user's requested claims or framing, unsolicited content review, unsolicited disclaimers, context leakage, visible internal-process leakage, claim-destination mismatch, delivery-model mismatch, source-relative commercial regression, or a material omission from the requested artifact.
 
 ## Evidence register
 

package/skills/agora/references/agora-publication.md

@@ -0,0 +1,137 @@
+# Agora publication privacy and provenance
+
+Use this reference only for an explicit request to inspect a local artifact before publication or external sharing. Agora remains a writing system during ordinary drafting.
+
+## Contents
+
+- [Purpose and boundary](#purpose-and-boundary)
+- [Model-level text watermark boundary](#model-level-text-watermark-boundary)
+- [Trigger and non-trigger rules](#trigger-and-non-trigger-rules)
+- [Run the audit](#run-the-audit)
+- [Interpret the report](#interpret-the-report)
+- [Format coverage](#format-coverage)
+- [Unicode review](#unicode-review)
+- [Metadata and privacy review](#metadata-and-privacy-review)
+- [Provenance review](#provenance-review)
+- [Publication decision](#publication-decision)
+- [Sources](#sources)
+
+## Purpose and boundary
+
+Inspect the artifact that will leave the user's control. Report hidden text controls, document metadata, review material, and provenance carriers without changing source bytes.
+
+Never describe this audit as a watermark remover, AI detector, authorship detector, humanizer, anonymizer, metadata cleaner, or proof that a file is safe or clean. Do not promise detector evasion. A missing finding means only that the named check did not find its configured signal.
+
+Claude's documented text marking is model-level. Do not imply that a Unicode scan can detect or remove it. Unicode review and file metadata review solve narrower publication-hygiene problems.
+
+## Model-level text watermark boundary
+
+A token-distribution watermark is introduced during generation by changing token-selection probabilities. It is not a hidden Unicode string or an ordinary file property. Detection can depend on the matching tokenizer, keyed hash or partition rule, normalization procedure, statistical threshold, and enough eligible text. Without the provider's verifier or equivalent configuration, the result remains `UNKNOWN`.
+
+The Kirchenbauer et al. green-list method is one research example, not evidence that Claude uses that exact scheme. It shows why short or low-entropy text can provide weaker statistical evidence, why editing may dilute rather than deterministically remove a mark, and why a detector can normalize whitespace, homoglyph, or zero-width changes before testing. Unusual Unicode is therefore neither evidence of a model-level watermark nor reliable evidence that one was removed.
+
+Do not estimate model-level watermark status from prose style, generic AI-detector scores, ad hoc word frequencies, or the absence of metadata findings.
+
+## Trigger and non-trigger rules
+
+Load this reference when the user explicitly asks Agora to:
+
+- inspect one or more local files before publication, submission, delivery, or external sharing;
+- find hidden Unicode, metadata, comments, notes, tracked changes, personal information, or provenance records;
+- perform a publication privacy or provenance audit;
+- review an artifact produced by another document, presentation, PDF, site, design, or image tool before release.
+
+Do not load or run the audit merely because writing is public, indexable, generated with AI, or being optimized for GEO/AEO. Do not run it against ordinary chat text unless the user supplies or identifies a local file. Do not create SVG, PNG, JPEG, PDF, DOCX, or PPTX files solely to make this workflow applicable.
+
+## Run the audit
+
+Use the shipped read-only script:
+
+```sh
+node <skill-root>/scripts/publication-audit.mjs <path...>
+```
+
+Use JSON when another tool or agent will consume the report:
+
+```sh
+node <skill-root>/scripts/publication-audit.mjs <path...> --json
+```
+
+Keep sensitive metadata values redacted by default. Use `--show-values` only when the user explicitly needs exact values. Use `--include-paths` only when full local paths are necessary. Use `--verify-c2pa` only when local C2PA verification is requested; the script supplies settings that disable remote-manifest fetching.
+
+The script may write an audit report only through an explicit `--output <path>`. It refuses to overwrite an existing report. It never writes to a source file.
+
+## Interpret the report
+
+Use the exact states:
+
+| State | Meaning |
+|---|---|
+| `FOUND` | The named check observed its configured signal. |
+| `NOT_FOUND_BY_THIS_CHECK` | The named check ran and did not observe that signal. |
+| `UNKNOWN` | Coverage was unavailable, unsupported, skipped, partial, or not requested. |
+| `ERROR` | The named check could not complete. |
+
+Use finding severities as routing aids, not verdicts:
+
+| Severity | Treatment |
+|---|---|
+| `info` | Normal technical detail that may still matter to publication. |
+| `review` | Inspect purpose and intended audience before publishing. |
+| `sensitive` | Likely personal, organizational, editorial, or hidden review material. |
+| `provenance` | Origin or processing information. Presence is not automatically harmful. |
+
+Never collapse `UNKNOWN` into success. Never translate `NOT_FOUND_BY_THIS_CHECK` into `clean`, `AI-free`, `human-written`, `anonymous`, or `safe`.
+
+## Format coverage
+
+| Format | Built-in inspection |
+|---|---|
+| TXT, Markdown | UTF-8 decoding and configured invisible or control characters. |
+| HTML | Text controls, selected metadata fields, and published comments. |
+| SVG | Text controls, metadata blocks, comments, and selected editor fields. |
+| DOCX | Core, application, and custom properties; comments; tracked changes; custom XML. |
+| PPTX | Core, application, and custom properties; comments; authors; notes; hidden slides; custom XML. |
+| PDF | Selected document-information fields, XMP presence, and C2PA carrier hints. |
+| PNG | Text chunks, EXIF-container presence, and C2PA carrier hints. |
+| JPEG | EXIF, XMP, IPTC, comment-container presence, and C2PA carrier hints. |
+
+ExifTool, when already installed, adds broader read-only metadata inspection for PDF, PNG, and JPEG. The script does not install it. Embedded Office media is reported as uninspected instead of being silently treated as clean.
+
+## Unicode review
+
+Review code point, count, and location. Do not delete characters by category.
+
+Zero-width joiners, zero-width non-joiners, variation selectors, and direction controls can be legitimate in emoji, Persian, Arabic, Hebrew, and other writing systems. Treat them as context-dependent. Treat directional overrides, internal byte-order marks, soft hyphens, and unexplained invisible separators as review findings, not proof of malicious or AI-generated text.
+
+## Metadata and privacy review
+
+Prioritize GPS coordinates, device or owner serials, email addresses, author and company names, local paths, comments, speaker notes, hidden slides, tracked changes, custom properties, custom XML, and edit history.
+
+Report field names and counts without values by default. Avoid pasting a full audit report into a hosted model when the report contains values the user has not approved for disclosure.
+
+## Provenance review
+
+Treat creator software, timestamps, XMP, EXIF, IPTC, and C2PA as provenance signals. Do not assume they are private, harmful, accurate, complete, or forged.
+
+C2PA can bind assertions and ingredients to an asset. A carrier hint is not cryptographic validation. `c2patool` validation may distinguish absent, reported, or problematic manifest information, but it cannot establish the full real-world truth of every assertion.
+
+Do not remove or invalidate provenance because it reveals AI involvement. Preserve, disclose, or remove metadata only through a separate user-directed editing workflow with exact fields and consequences identified. This audit provides no cleaning operation.
+
+## Publication decision
+
+Return findings first, then coverage gaps, then source-integrity status. Separate three decisions:
+
+1. Privacy: whether identified values or hidden review material should be disclosed.
+2. Provenance: whether origin and processing records should be preserved or disclosed.
+3. Integrity: whether the inspected source stayed byte-identical.
+
+When remediation is requested later, name exact fields and create a separate copy. Never modify the only source artifact. Re-audit the derived copy and report any provenance invalidation or coverage gap caused by the edit.
+
+## Sources
+
+- Anthropic, [How Claude marks AI-generated content](https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content)
+- Kirchenbauer et al., [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226) (research mechanism example, not a Claude implementation specification)
+- Coalition for Content Provenance and Authenticity, [C2PA technical specification](https://spec.c2pa.org/specifications/specifications/2.4/specs/ContentCredentials.html)
+- Content Authenticity Initiative, [Using C2PA Tool](https://github.com/contentauth/c2pa-rs/blob/main/cli/docs/usage.md)
+- Phil Harvey, [ExifTool application documentation](https://exiftool.org/exiftool_pod2.html)