/ad-script and /hook when the expert's niche is health or wellnessThis skill is a niche addendum — it does not replace ad-hook-generation, it extends it. When the expert's offer touches health or wellness (weight loss, hormone health such as polycystic ovary syndrome or PCOS, chronic conditions, energy, fitness, longevity, or mental health as a physical symptom), twelve additional frameworks labelled H1 through H12 become available. The hook generation step selects 2–3 of these and combines them with 1–2 frameworks from the core ad-hook-generation library. The result is a hook set shaped to a health audience's specific psychology: high shame, many failed attempts, high scepticism, and a body relationship that sits at the centre of identity. This skill fires inside the /ad-script hook chain (after ad-hook-principles and ad-hook-generation) and inside the standalone /hook command. It does not run for non-health offers. Health claim compliance is a structural sensitivity throughout — two frameworks (H3 Medical Contrast, H11 Quiet Desperation) carry elevated Meta policy risk and link downstream to ad-compliance.
Health-vertical hooks carry a higher rate of Meta ad rejection than other niches. Two frameworks in particular — H3 Medical Contrast (what your doctor won't test for) and H11 Quiet Desperation (if you've reached the point where...) — require a compliance review before the hook is used in a live ad. H3 hooks framed as "doctors lie" or "they don't want you to know" are a rejection trigger; frame as an information gap instead. H11 hooks that imply the product is the only option or invoke fear of harm are also flagged. All H3 and H11 hooks should pass through ad-compliance before production. Do not combine H3 with an explicit disease-state claim in the same hook. Claim substantiation — the evidence backing any health claim — is required for any hook that names a mechanism, a test, or a physiological process.
All calls go through OpenRouter and are logged with model and cost against the parent agent_run row. The deterministic-first rule applies: counting, scanning, and threshold checks are plain code, never a model.
| Task | Model | Why |
|---|---|---|
Fetching Supabase views: expert_profile, active_launch_learnings_for_[expert], winning_ads filtered to health vertical | Plain code — no language model | Deterministic data fetch. A model must never decide whether a required view exists or is empty. If expert_profile has no row with status = 'current', the skill halts — plain code makes that call, not inference. |
Avoid/forbidden phrase scan against active_launch_learnings_for_[expert] | Plain code — no language model | String matching is deterministic. Whether a hook angle contains a forbidden term is a pass/fail code check, not an LLM judgment call. Runs before any generation step. |
| Framework selection — choosing 2–3 H-frameworks given the condition, demographic, and launch learnings signals | claude-sonnet-4-6 (Sonnet 4.6) via OpenRouter | Bounded selection task with a fixed 12-framework vocabulary. The model is choosing from a labelled list, not generating open-ended structure. Sonnet 4.6 is sufficient for this level of reasoning at lower cost than Opus 4.8. |
| Hook generation — writing final hook lines from selected H-frameworks, the Voice Book, and the demographic profile | claude-opus-4-8 (Opus 4.8) via OpenRouter | Voice fidelity is a synthesis task. Every health hook must earn trust with a sceptical, high-shame audience. The language must feel like the expert, not like a copywriter. Opus 4.8 is the floor for generation — a weaker model produces generic health-ad language that fails the Voice Book fidelity check. |
| Variation expansion — generating a controlled variation set (6 or more hook variants on the same framework) across message, visual, format, and borrowing dimensions | gpt-5.5 via OpenRouter | Broad, controlled variation is GPT-5.5's strength for high-volume batches. Used only when the operator explicitly requests a variation batch of 6 or more hooks from a single framework. |
First-time health hook batch for a new expert with no winning_ads data to cross-check (e.g. Elena or Alba in early setup) | claude-fable-5 (Fable 5) via OpenRouter — opt-in only, operator sign-off required | Ceiling one-shot job with no internal reference data to validate against. Costs 2× Opus 4.8 and carries a mandatory 30-day data retention obligation. Requires written customer personal-data approval before any expert voice material is included in the input. Never a default — the operator must explicitly request it. |
The skill source specifies Sonnet 4.6 (framework selection) and GPT-5.5 (variation expansion). The system-wide context pack sanctions three model tiers: plain code, Opus 4.8, and Fable 5 (ceiling, opt-in). Sonnet 4.6 and GPT-5.5 are not in that list. The skill source governs what the skill actually does — so this dossier documents both models as-built. The gap needs operator sign-off: either the context pack's model vocabulary expands to include Sonnet 4.6 and GPT-5.5 for bounded health-niche tasks, or the skill is updated to route framework selection to plain code and variation expansion to Opus 4.8. Until resolved, treat this as an open configuration item.
What goes in, how the skill runs, and what comes out.
| Input | What it is |
|---|---|
Parent run_id | The agent_run row ID from the enclosing /hook or /ad-script invocation. The skill halts immediately if this is absent — it does not open its own agent_run row, it logs all model calls against the parent. |
expert_profile (Supabase Layer 2 — the Voice Book) | The expert's Voice Book row from Supabase, filtered to status = 'current' and the expert's namespace (e.g. elena, alba). Contains voice signature patterns, authentic vocabulary, forbidden tones, and the ICP (ideal customer profile — the specific person the expert's offer is designed for) detail used to calibrate every hook line. The skill halts if no current row exists — health hooks require a confirmed Voice Book to meet voice fidelity requirements. |
active_launch_learnings_for_[expert] (Supabase Layer 5) | The current avoid, forbidden, and positive pattern rows from ad-autopsy runs for this expert. The avoid and forbidden rows are scanned by plain code before any hook generation begins — any hook angle that contains a forbidden topic is blocked. Positive rows reinforce framework selection: proven H-frameworks from prior batches are weighted first. |
winning_ads view (Supabase Layer 4) filtered to health vertical | The materialised view of ads with a winning or performing status for the expert, filtered to ads tagged with the health vertical. Tells the skill which H-frameworks are already proven for this expert — proven frameworks are prioritised before untested ones. |
| Condition / health struggle (from operator brief) | The specific health condition or struggle the ad is targeting. Examples: PCOS (polycystic ovary syndrome — a hormone condition affecting ovulation and metabolism), acne, perimenopause, gut dysbiosis, chronic fatigue. Narrows H-framework selection to the most relevant subset. |
Demographic profile (from ad-audience-language-dna) | The demographic language calibration layer — age bracket, emotional arc, intensity level, vocabulary preferences — passed in from ad-audience-language-dna. Shapes how the selected H-frameworks are written (e.g. a 45-year-old woman with PCOS needs different language intensity than a 28-year-old with acne). |
| Cold / warm audience signal | Whether the hooks are for top-of-funnel (TOF — cold traffic, people who have never seen the expert) or retargeting (warm, people who have engaged before). Cold hooks prioritise frameworks that meet the viewer in their current physical experience (H1, H4, H9). Warm hooks can reference earlier content and layer in sustainability promises (H10) or "finally" permission (H5). |
/ad-script hook chain output or the standalone /hook outputEach health hook follows the same output format as ad-hook-generation — framework label, format, visual direction, audio line, and "why it stops the right viewer" rationale. The H-framework ID replaces or supplements the F/HT/W framework ID from the core library.
| Field | What it means |
|---|---|
| Framework | The H-framework ID and name — e.g. "H4 — Condition Callout". Where the health framework is combined with a core framework from ad-hook-generation, both IDs appear (e.g. "H2 + F1 — Root Cause Mechanism × Contrarian Claim"). |
| Format | The video format — talking head (DTC — direct to camera), text-on-screen/voiceless, skit/scene, or UGC (user-generated content style). Mute-first compliance requires at least 2 hooks in every batch of 10 to be fully text-on-screen/voiceless. Format recommendations come from the style system and the operator's brief. |
| Visual | Shot direction at the level needed by a camera operator — framing, setting, any on-screen text overlays, timing of their appearance. Does not include post-production detail (colour grading, motion graphics) — that lives in the production brief. |
| Audio | The exact spoken hook line, written in the expert's voice as confirmed by the Voice Book. For voiceless formats, this field contains the on-screen text sequence and beat timing instead. |
| Why it stops the right viewer | A plain-English rationale explaining which specific phrase or structural element creates the pattern interrupt, what psychological mechanism it activates (open loop, mirror language, shame release, etc.), and why it qualifies the right viewer while repelling the wrong one. This is the reviewer's primary audit surface. |
| Compliance flag | Present only on H3 and H11 hooks. States the specific policy risk and the required action — either reframe the language or route through ad-compliance before use. Empty for all other H-frameworks. |
| ID | Name | Core structure | Best used when |
|---|---|---|---|
| H1 | Symptom Recognition | "If you [specific physical symptom], your body is trying to tell you something." Meets the viewer in their current physical experience. No promise required — just recognition. | The audience can name a specific physical sensation they experience regularly. Best for TOF cold traffic where trust is zero and connection is everything. |
| H2 | Root Cause Mechanism | "The real reason [symptom / struggle] has nothing to do with [common blame]." Releases internalized blame — from doctors, from diet culture, from themselves — and opens curiosity about what the real cause is. | The audience has been told (or tells themselves) that their struggle is caused by willpower failure, laziness, or age. Combines naturally with F1 (Contrarian Claim) from the core library. |
| H3 | Medical Contrast | "What your doctor won't test for / didn't tell you / missed." Validates the experience of being dismissed by standard medical care. Elevated compliance risk — see compliance callout above. | The audience has received "normal" blood work results but still feels unwell. Frame as an information gap in standard care, never as medical system conspiracy. Route all H3 hooks through ad-compliance before use. |
| H4 | Condition Callout | "[Condition]? There's one thing connecting [symptom cluster]." Condition naming = instant relevance filter for diagnosed populations. Works because it stops only the people who already know they have that condition. | The audience is already diagnosed or self-diagnosed. Best for PCOS (polycystic ovary syndrome), Hashimoto's thyroid disease, insulin resistance — conditions with active online communities. Combines naturally with F5 (Identity Callout) from the core library. |
| H5 | "Finally" Permission | "After [years / attempts] of [struggle], there might be something you haven't tried." Validates the journey. "Might be" language reduces skepticism — it does not claim certainty. | The audience has a documented history of trying and failing. Best for retargeting warm audiences who have engaged with content before. Works well after the expert has established trust through organic content. |
| H6 | Body Signal Reframe | "Your [symptom] isn't the problem — it's your body trying to [message]." Shifts from an adversarial body relationship to a curious partnership. Reduces shame by reframing the symptom as communication. | The audience has been fighting their body. Best for cravings, fatigue, and weight — symptoms that carry heavy self-blame. Combines naturally with F7 (Paradox Hook) from the core library. |
| H7 | Anti-Diet / Anti-Hustle | "What if [result] didn't require [punishment method]?" Health audiences are exhausted from extreme approaches. Signals a different, less punishing path. | The audience has tried extreme methods — severe calorie restriction, elimination diets, daily high-intensity training. Names a specific punishment approach they recognise from their own history. |
| H8 | Age / Stage Specificity | "[Result] looks different after [age / stage] — here's what actually works." Generic advice fails for specific life stages. Creates authority through the specificity of naming a stage with genuinely different requirements. | The audience is in perimenopause, post-40, or post-partum — stages where standard advice demonstrably fails. Names the stage, not just the age. |
| H9 | Invisible Struggle | "When [invisible symptom] and no one understands what you're going through." Many health struggles — fatigue, brain fog, hormone dysregulation — are invisible. Being seen is a powerful emotional entry point. | The audience looks "fine" externally but feels terrible internally. Works best when the expert can speak from lived experience or a student's experience that the audience will immediately recognise as their own. |
| H10 | Sustainability Promise | "[Result] that actually lasts — without [unsustainable method]." The core fear in health is not "will this work?" — it is "will this last?" Names both the result and the sustainability mechanism in the same hook. | The audience has experienced results that did not hold. Best for retargeting and warm audiences who have heard promises before. Pair with believable numbers from the ad-scripter-reference vertical ranges. |
| H11 | Quiet Desperation | "If you've reached the point where [desperation marker], this is for you." Names the depth of the struggle. Elevated compliance risk — must be paired with a constructive promise in the script body. See compliance callout above. | The audience is in a late-stage, high-intensity struggle where they have exhausted standard options. Must be followed by genuine hope. Never exploit the desperation — validate and redirect. |
| H12 | The One Thing | "The ONE thing that [addresses multiple symptoms]." Health audiences are overwhelmed by complexity. "One thing" creates relief and curiosity simultaneously — it reduces the cognitive load of a category crowded with multi-step protocols. | The audience has been given complex, multi-step protocols that didn't stick. The expert's mechanism genuinely addresses multiple symptoms through a single lever. The "one thing" must be real — do not use this framework if the mechanism is actually multi-step. |
Items 1, 3, 6, and 7 are deterministic — plain code scans for word/phrase patterns. Items 2, 4, and 5 require model review. The code scan runs before the model review step.
| Check | What to verify | How |
|---|---|---|
| 1. Shame-aware | No blame language. No "finally get serious", "stop making excuses", or equivalent. The hook must not imply the viewer caused their own struggle through weakness. | Plain code — word/phrase scan |
| 2. Sustainability | Results are framed as lasting, not as a fast transformation. If a Time Collapse hook (from ad-hook-generation) is used, a sustainability caveat is required. | Model review |
| 3. Believable numbers | Any numbers used fall within the vertical-specific believable ranges in ad-scripter-reference Part 2. Health numbers outside the reference ranges require operator sign-off before use. | Plain code — range check against reference table |
| 4. Failed attempts acknowledged | The hook shows it understands the audience has tried before. Hooks that assume a naive first-time buyer will not resonate with health audiences. | Model review |
| 5. Body-positive framing | The hook frames the body as a system to work with, not an adversary to defeat. No language that positions the body as broken, failed, or the enemy. | Model review |
| 6. No medical overreach | No language that sounds like diagnosis, treatment recommendation, or cure claim. The skill addresses symptoms and mechanisms — never disease states. | Plain code — phrase scan for diagnostic and treatment language |
| 7. Launch learnings applied | No topics from the avoid or forbidden sections of active_launch_learnings_for_[expert] appear in any hook angle. | Plain code — scan against live Supabase rows before generation |
Pending: Great Ad rubric consolidated V1 — when the official rubric is released, this checklist will be extended with rubric criteria specific to health-vertical hooks. Do not add rubric criteria before the official release.
No health-niche eval run exists yet (Elena and Alba proof banks are in setup). This is a representative example, constructed from the Elena/PCOS config and the H-framework structures above. It follows the exact output format of the real eval artifacts in evals/runs/2026-06-10-depth-eval/hg-cold/.
HOOK BATCH: Elena — PCOS / Hormone Health
TARGET: Women 28-42 with PCOS, insulin resistance, or stubborn weight despite "doing everything right"
AWARENESS: Problem-Aware (know the struggle; unaware of Elena's specific approach)
FUNNEL: TOF / COLD | AUDIENCE SIGNAL: Diagnosed or self-diagnosed PCOS, failed diet attempts
MECHANISM: TEASER — never name phases or steps
---
HOOK 1
Framework: H4 — Condition Callout
Format: Talking head (Elena, direct to camera — natural home setting, lo-fi)
Visual-intent: Elena already mid-sentence in a warm, real setting — no logo, no intro — lower-third text "PCOS and can't lose weight no matter what?" fades after 2 seconds.
Mute-ready: no
Reveal: TEASER
"PCOS, insulin resistance, and stubborn belly fat — if that's you, they're all connected. And it's not your diet. It's not your willpower. There's one piece of this nobody explained to you."
Why it stops the right viewer: Naming PCOS in the first word is a precision relevance filter. Only the right person stays. "It's not your diet. It's not your willpower." lands as shame release — the viewer has been blaming themselves for years. "One piece nobody explained" creates an open loop without revealing the mechanism. The TOF viewer has to keep watching.
---
HOOK 2
Framework: H2 + F1 — Root Cause Mechanism × Contrarian Claim
Format: Text-on-screen / voiceless (mute-first)
Visual-intent: Dark background, no face, no branding — three text beats in sequence: "The real reason you can't lose weight with PCOS..." → "...has nothing to do with eating less." → "Here's what's actually happening."
Mute-ready: yes
Reveal: TEASER
VOICELESS — on-screen text only. Optional: quiet ambient sound.
Why it stops the right viewer: Beat 2 violates the viewer's assumption — she has been eating less, and it hasn't worked. The contrarian reframe ("nothing to do with eating less") validates her experience without shaming her for trying the wrong thing. Beat 3 is pure open loop — the viewer has to keep watching to learn what IS actually happening. Mute-first compliance satisfied.
---
HOOK 3
Framework: H9 — Invisible Struggle
Format: Talking head (Elena, confessional tone — close to camera, low production feel)
Visual-intent: Elena close to the lens, personal and unscripted — text overlay "When no one believes you." fades in briefly, then disappears as she continues.
Mute-ready: no
Reveal: TEASER
"When your blood work comes back 'normal' and your doctor says you're fine — but you're exhausted, you can't lose weight, and you know something is wrong. I see you. And there's a reason for all of it."
Why it stops the right viewer: "Blood work comes back normal" is the single most common PCOS experience before diagnosis — the viewer hears her exact situation in the first 4 seconds. "I see you" is an explicit emotional acknowledgment of the invisibility. "There's a reason for all of it" creates an open loop: what is the reason? The hook does not exploit the desperation — it validates and then points toward resolution. This is the warmest hook in the batch and will qualify a viewer who has been dismissed repeatedly by the medical system.
---
HOOK 4
Framework: H6 — Body Signal Reframe
Format: Talking head (Elena, measured, warm — slightly slower pace)
Visual-intent: Elena at a desk or kitchen table, looking directly into camera — text overlay "Your body isn't broken." holds for 2 seconds, then disappears as she speaks.
Mute-ready: no
Reveal: TEASER
"Your cravings aren't weakness. Your fatigue isn't laziness. Your body is sending you a signal — and when you learn to read it, everything changes."
Why it stops the right viewer: "Your body isn't broken" directly addresses the core fear of the PCOS viewer — the belief that her body is fundamentally defective. "Cravings aren't weakness" and "fatigue isn't laziness" each release a specific shame the audience carries. The phrase "learn to read it" creates curiosity without revealing the mechanism. No urgency, no hype — just a reframe that makes the viewer feel understood.
---
COMPLIANCE NOTE (Hook 3 only): H9 used with medical validation gap context. Framed as an information gap ("there's a reason for all of it") not as medical system conspiracy. No disease-state claim. No treatment claim. This hook does not require ad-compliance review — H3 and H11 flags are not triggered. If the script body uses any H3-type language about specific tests, run the full hook + body through ad-compliance before production.
FRAMEWORK DISTRIBUTION: H4, H2+F1, H9, H6 — four distinct frameworks, three H-frameworks from this skill, one core combination (H2+F1). No framework repeats.
MUTE-FIRST COMPLIANCE: Hook 2 is fully voiceless. One-hook minimum for a short batch satisfied.
SHAME-AWARE SCAN: Passed. No blame language, no "finally get serious", no "stop making excuses" variants detected.
MEDICAL OVERREACH SCAN: Passed. No diagnosis language, no treatment claims, no disease-state assertions.
LAUNCH LEARNINGS SCAN: Elena active_launch_learnings — no avoid/forbidden topics present in any hook angle. (Pending: Elena proof bank in setup — no positive signals available to reinforce framework selection yet. First-batch default: framework selection by condition + demographic only.)
| Term | In full / what it means |
|---|---|
| H1 | Symptom Recognition — Hook framework 1 in this health addendum. Meets the viewer in a specific physical symptom they experience regularly. No promise, just recognition. |
| H2 | Root Cause Mechanism — Hook framework 2. "The real reason [symptom] has nothing to do with [common blame]." Releases internalized blame and opens curiosity. |
| H3 | Medical Contrast — Hook framework 3. "What your doctor won't test for / didn't tell you." Validates the experience of being dismissed by standard medical care. Elevated Meta compliance risk — requires ad-compliance review before use. |
| H4 | Condition Callout — Hook framework 4. Names the specific diagnosed or self-diagnosed condition in the first words of the hook. Instant relevance filter for diagnosed populations. |
| H5 | "Finally" Permission — Hook framework 5. "After [years / attempts] of [struggle], there might be something you haven't tried." Validates the journey without over-promising. |
| H6 | Body Signal Reframe — Hook framework 6. "Your [symptom] isn't the problem — it's your body trying to tell you something." Shifts from adversarial body relationship to curious partnership. Reduces shame. |
| H7 | Anti-Diet / Anti-Hustle — Hook framework 7. "What if [result] didn't require [punishment method]?" Signals a less punishing path for audiences exhausted by extreme approaches. |
| H8 | Age / Stage Specificity — Hook framework 8. "[Result] looks different after [age / stage]." Creates authority through naming a life stage with genuinely different physiological requirements. |
| H9 | Invisible Struggle — Hook framework 9. "When [invisible symptom] and no one understands." Addresses health struggles that are invisible externally but debilitating internally. Being seen is the emotional entry point. |
| H10 | Sustainability Promise — Hook framework 10. "[Result] that actually lasts — without [unsustainable method]." Addresses the core health-audience fear: not "will this work?" but "will this last?" |
| H11 | Quiet Desperation — Hook framework 11. "If you've reached the point where [desperation marker], this is for you." Names the depth of struggle. Elevated Meta compliance risk — must be paired with a constructive promise. Requires ad-compliance review before use. |
| H12 | The One Thing — Hook framework 12. "The ONE thing that [addresses multiple symptoms]." Provides relief from the complexity of a crowded health category. The "one thing" must be genuinely multi-symptom in its mechanism — do not use if the approach is actually multi-step. |
| PCOS | Polycystic Ovary Syndrome — A hormone condition affecting ovulation and metabolism, common in women of reproductive age. Causes irregular periods, insulin resistance, and difficulty losing weight. The primary health condition for Elena's expert niche in this system. |
| TOF | Top of Funnel — Cold traffic — people who have never seen the expert's content before. Health TOF hooks must meet the viewer in their current physical experience because trust is zero and the viewer has no prior relationship with the expert. |
| MOF | Middle of Funnel — Warm traffic — people who have seen content or engaged with an ad but have not yet taken a conversion action. H5 (Finally Permission) and H10 (Sustainability Promise) perform best here. |
| BOF | Bottom of Funnel — Hot traffic — people who have seen the free training or visited the booking page but have not yet converted. High-intent retargeting. |
| CPBC | Cost Per Booked Call — How much was spent on Meta ads to get one person to book a sales call with the expert. The primary success metric for health experts running high-ticket cold traffic campaigns. |
| CPL | Cost Per Lead — How much was spent to get one person to register for a free training or lead magnet. Used as an early-signal proxy when CPBC data is not yet available (e.g. Elena's first 30 ads). |
| ICP | Ideal Customer Profile — The specific person the expert's offer is designed for, described in enough detail that every hook and script can be calibrated to their language, fears, and desires. Lives in the Voice Book's expert_profile row in Supabase. |
| Voice Book | Expert Voice Book — Layer 2 of the data layer. The structured voice profile for an expert, built by ad-voice-profile and stored as the expert_profile row in Supabase with status = 'current'. Contains voice signature patterns, authentic vocabulary, forbidden tones, and ICP detail. Previously called "Layer 3 calibration" — that name is deprecated. This skill halts if no current Voice Book row exists for the expert. |
| Hook rate | Hook rate — The percentage of viewers who watch past the opening hook section of an ad (typically the first 3–5 seconds on Meta). A high hook rate means the hook stopped the scroll; it does not mean the ad converted. A high hook rate with low conversion is a signal that the hook qualified the wrong viewer. |
| Claim substantiation | Claim substantiation — The evidence backing a health claim. On Meta, any hook that names a physiological mechanism, a test, or a specific health outcome requires that the claim is grounded in legitimate science and that the expert can substantiate it if challenged. Unsubstantiated health claims are a primary cause of Meta ad rejection in the health vertical. All H3 hooks (Medical Contrast) require substantiation for any clinical framing used. |
| DTC | Direct to Camera — A video ad format where the presenter speaks directly into the camera lens, creating a one-on-one feel. The primary format for most of Elena's and Alba's talking-head hooks. |
| UGC | User-Generated Content — An ad format that mimics organic, creator-style content — informal, handheld, less polished. Used in health niches to reduce the "ad" feeling that triggers skepticism in high-shame audiences. |
| PII | Personally Identifiable Information — Any data that can identify a specific individual — names, email addresses, call recordings. Fable 5 carries a mandatory 30-day data-retention obligation; PII must never be included in Fable 5 inputs without written customer approval. Health data (conditions, symptoms, diagnoses) is sensitive PII and is scrubbed at the raw-to-extract boundary. |
| HITL | Human in the Loop — The operator approval gate. In V1, no creative moves to production without the operator reviewing and approving the hook batch output. The system captures both the raw hook output and the operator's edited final — the delta is the agentic performance signal. |
| Hermes | Hermes (autonomous agent layer) — The always-on agent system that maintains the Supabase data layer. Voice-Hermes maintains the Voice Book rows. Curator-Hermes maintains the competitor and swipe library layers. This skill reads from Hermes-maintained views; it does not trigger Hermes scrapes. |
| Supabase | Supabase (database platform) — The PostgreSQL-based database hosting the SKL ad creative system data layer. This skill reads three Supabase views at runtime: expert_profile, active_launch_learnings_for_[expert], and winning_ads. Agents read Supabase views — never Fibery, Drive, or repo files at runtime. |
| Fibery | Fibery (project management and system of record) — The system of record for all creative entities and their production status. The operator's primary work surface in the creative pipeline. Hermes agents sync Fibery data into Supabase views; this skill reads the views, not Fibery directly. |
| OpenRouter | OpenRouter (model routing proxy) — The API gateway through which all language model calls in this system route. Every call logs model ID, provider, and cost to the parent agent_run row. |
| F1–F12 | Cold Traffic Core Hook Frameworks 1 through 12 — The 12 core cold-traffic frameworks in ad-hook-generation: Contrarian Claim, Unexpected Outcome, Hidden Truth, Specific Transformation, Identity Callout, Mistake Reveal, Paradox Hook, Question Interrupt, Time Collapse, What If Reframe, Social Proof Lead, Native Story Opening. H-frameworks from this skill combine with F-frameworks — e.g. H2 (Root Cause Mechanism) maps naturally to F1 (Contrarian Claim); H6 (Body Signal Reframe) maps to F7 (Paradox Hook). |
| HT1–HT12 | High-Ticket Cold Traffic Hook Frameworks 1 through 12 — The 12 high-ticket cold-traffic frameworks in ad-hook-generation. Elena's offer is high-ticket; HT-frameworks are used alongside H-frameworks in her hook batches. |
| W1–W10 | Warm Traffic Retargeting Frameworks 1 through 10 — The 10 warm-traffic hook frameworks in ad-hook-generation. Relevant for health retargeting campaigns; H5 (Finally Permission) and H10 (Sustainability Promise) function as warm-traffic complements. |
| ad-compliance | ad-compliance (downstream reactive skill) — Phase 5 skill that diagnoses and fixes Meta-rejected ads. H3 (Medical Contrast) and H11 (Quiet Desperation) hooks must be reviewed against the ad-compliance checklist before production. Ad-compliance is reactive in V1 — it fires after rejection. Pre-production compliance review for H3 and H11 is the preventive step that reduces rejection rate. |
| Elena | Elena (PCOS / hormone health expert) — One of the SKL expert identities. Expert namespace: elena. Her niche is polycystic ovary syndrome and hormone health. Status: 30 ads scripted, proof bank in setup, inactive in cron until Fibery performance data is populated. The primary health-niche expert for whom this addendum was designed. |
| Alba | Alba (acne / skin health expert) — One of the SKL expert identities. Expert namespace: alba. Her niche is acne and skin health. Status: setup phase, proof bank not yet built. A secondary health-niche expert for whom H4 (Condition Callout) and H6 (Body Signal Reframe) are expected to be most relevant. |