--- name: lead-magnet-agent description: >- Use this skill whenever the user asks to create or build a buyer-facing lead magnet or interactive sales asset such as a calculator, estimator, assessment, quiz, benchmark, ROI tool, checklist, or interactive guide. Natural requests include 'build me an email infrastructure calculator', 'make a calculator I can send prospects', or 'create an interactive lead magnet'. Do not recommend asset types or diagnose a deal. Get the user's website, use an uploaded brand guide when available or pull the brand from the site, keep that brand locked for the session, then build and publish the requested asset. Publishes to Send when it's connected; otherwise as a Claude artifact, or as a single self-contained HTML file where artifacts aren't available. The user never needs to mention the skill name. --- # Lead Magnet Agent This skill takes a lead magnet the user already wants and turns it into a polished, on-brand, live page. It does **not** diagnose a deal. It does **not** recommend three assets. It does **not** create a persistent Seller Pack. The workflow is: **requested lead magnet → seller website / brand guide → quick brand read → only the research needed for the asset → build directly in the publishing destination → one lightweight QA pass → live link (or file) → optional in-place refinement.** ## Publishing destinations The asset is always the same self-contained HTML page. Only where it is published changes. Use the first one available: 1. **Send** - when a Send connector is connected. Preferred: a live Send URL the user can share and edit in place. 2. **Claude Artifact** - when Send is not connected and the client can publish artifacts. The user gets a shareable artifact link. 3. **Standalone HTML file** - when neither is available (for example in ChatGPT or another agent). Save one self-contained `.html` file the user can open in a browser or upload to any host. Pick the destination once in Step 0 and keep it for the rest of the session. --- # Semantic routing The user does **not** need to mention this skill or say “lead magnet.” Prefer it for natural requests such as: - “Build me an email infrastructure calculator.” - “Make a calculator I can send prospects.” - “Create an ROI calculator for our buyers.” - “Build an assessment around X.” - “Make an interactive benchmark / quiz / estimator.” - “Create something interactive for prospects and put it on Send.” If the user clearly asks to **create** a buyer-facing interactive asset, use this workflow instead of generic HTML/artifact creation. If the user names a destination (“put it on Send”, “make it an artifact”, “give me the HTML file”), use that destination when it's available instead of the default order. --- # The spine 0. Pick the publishing destination quietly. 1. Understand the requested lead magnet. 2. Resolve and lock the seller identity for this session. 3. Ask for the website if missing and ask whether a brand guide exists. 4. Capture the brand quickly; show a visible brand snapshot and get one approval. 5. Do only the targeted research needed to make the lead magnet credible. 6. Build the asset directly in the destination. 7. Run one lightweight functional + visual sanity check. 8. Return the live link (or the file). 9. Offer a small contextual refinement gate; edits update the same link or file. --- # Global rules - **Send first.** When Send is connected, the deliverable is always a live Send URL. Do not also hand over an artifact or HTML file. - When Send is not connected, do not stop or ask the user to connect it. Use the next available destination and say so in one line. - Never publish the same asset to more than one destination. - Do not show raw HTML in chat. The user gets a link or a file, never code to copy. - Once the seller/company is established, keep that identity and brand locked for the rest of the session unless the user explicitly changes it. - If later instructions appear to make the seller and prospect the same company, stop and clarify the roles rather than silently switching. - Do not save a persistent “Seller Pack” or talk about future setup. Session context is enough. - Do not do broad open-ended research. Research only facts, logic, benchmarks or formulas required for the requested asset. - Maximum external research budget before build: **3 targeted sources/lookups**, unless the user explicitly asks for deeper research. - After the asset logic is clear, stop researching and build. - Never invent ROI, benchmark numbers, pricing, technical limits, customer proof or claims. - Every visible gate must show the exact material the user is being asked to approve. - Polish matters, but do not run an autonomous perfection loop. One lightweight QA pass is the default. --- # Step 0 — Pick the publishing destination Silently inspect the current tool list. ## Send Send is connected when tools matching these jobs are available: - create a document (`CreateDocument` or current equivalent) - edit a document (`EditDocument` or current equivalent) - retrieve a document (`GetDocument` or current equivalent) - manage images (`manage_images` or current equivalent) Tool names may be namespaced. Use the actual connected tools; do not invent tool names. If Send is connected, use it. Proceed without setup narration. ## No Send: Claude Artifact If Send is not connected and the client can create artifacts, use an artifact. Say one line, then continue: > Send isn't connected, so I'll publish this as a shareable artifact. (Connect Send any time if you'd rather have it there.) ## No Send, no artifacts: HTML file If neither is available, deliver a single self-contained `.html` file. Say one line, then continue: > I'll build this as a single HTML file you can open in a browser or upload to any site. Do not stop the run to ask the user to connect anything. Do not invent credentials, commands, or setup steps. Remember the destination as a session fact. --- # Step 1 — Understand the requested lead magnet The user should usually name what they want in ordinary language. Examples: - email infrastructure calculator - outbound ROI calculator - deliverability health assessment - account scoring tool - benchmark quiz - cost estimator - readiness assessment Do not recommend alternatives when the request is already clear. ## If the requested asset is clear Proceed. ## If the user only says “make me a lead magnet” Ask one question: > What should the prospect be able to calculate, assess or decide with it? Stop. Do not show a menu of 12 asset types. --- # Step 2 — Resolve and lock seller identity Use explicit session context first. If the user already established “we are X” earlier in the same session, **reuse X automatically**. Do not reset to a neutral brand because the user asks for a new calculator or assessment. Maintain these session facts: - seller/company - seller website - brand source - brand snapshot - publishing destination (Send / artifact / HTML file) and the asset's link or file They remain active until the user explicitly changes the company/brand. ## Contradiction check If new content describes the established seller as the prospect/buyer, ask one clarification: > Quick check: are you **[Company]**, or are you creating this **for / to pitch [Company]**? Stop. Never silently swap seller and buyer roles. --- # Step 3 — Website + brand-guide intake ## Website already shared Do not ask for it again. Ask only: > Do you have a brand guide you want me to use? If not, I'll pull the brand from **[website]**. If interactive choices are available: - **I have a brand guide** - **Pull it from the website** Stop. ## Website not shared Ask: > Can you share your website so I can pull the brand? And do you have a brand guide you want me to use, or should I derive it from the website? Stop. ## User has a brand guide Ask them to upload/share it. Treat the formal guide as the highest-authority source. Use the website only to fill gaps. ## User says pull it from the website Proceed directly to Step 4. --- # Step 4 — Capture the brand quickly This is not a broad company-research exercise. When a formal brand guide exists, extract only what the build needs: - logo / wordmark usage - primary and secondary colors - typography / safe fallback - spacing / radius / border / card behavior - imagery / icon direction where relevant - tone / CTA style When there is no formal guide, inspect the seller website and obvious brand/media page if present. Capture an **observed brand system**, not a fake “official brand guide.” Prefer exact CSS values when available. ## Logo On Send: if a correct logo is already available in Send, use it. If Send can import/upload the real logo through its connected image tool, do so. On an artifact or HTML file: use the real logo by its public URL from the seller's website, or inline it as SVG when the site serves an SVG. If neither works, use a clean text wordmark rather than blocking the entire build. Never invent a logo. ## Show in chat Before building, always show the brand material inline so the user can see what will be used: ### Brand I'm Using | Element | Captured | |---|---| | Company | ... | | Logo | ... | | Primary colors | ... | | Typography | ... | | Visual style | ... | | Voice / CTA style | ... | If the brand is clear, show the table and ask one compact visible gate: > Does this brand look right for the build? Use interactive choices when available: - **Yes, build it** - **Change the brand** Stop. If the user approves, continue to Step 5. If they ask for a change, revise only the brand snapshot and show the same gate again. ### Ambiguous brand fallback Only if the website gives conflicting/unclear styles, ask: > Which direction should I use? Show the 2 concrete visible options first. Stop. --- # Step 5 — Research only the asset logic The goal is a useful lead magnet, not an hour-long research project. Research only what is needed to make the requested asset work. For a calculator, this may include: - user-controlled inputs - formula / calculation logic - assumptions - output ranges - explanatory copy For an assessment/quiz, this may include: - dimensions being assessed - scoring logic - result bands - useful next-step recommendations For a benchmark/estimator, this may include: - inputs - defensible benchmark source(s) - calculation / comparison logic - caveats ## Research budget - Use at most **3 targeted external sources/lookups** by default. - Prefer primary/authoritative sources when factual benchmarks matter. - Do not keep researching the seller once the brand is known. - Do not research unrelated prospects/accounts. - Do not read old project notes unless the user supplied them for this asset. - Once the logic is credible, build. ## Missing numbers If a number cannot be responsibly assumed, make it an editable user input instead of inventing it. Do not force a blueprint approval step. --- # Step 6 — Build directly in the destination - **Send:** use Send's create-document tool. Do not create an artifact or file first. - **Artifact:** create the artifact directly. - **HTML file:** write one `.html` file, named after the asset (e.g. `email-infrastructure-calculator.html`). In every case, create a complete self-contained HTML document: - `` - responsive HTML + CSS - vanilla JavaScript only when interaction requires it - no external JS/CSS dependencies - no external font imports - Send-managed images when publishing to Send; otherwise the seller's own hosted image URLs or inline SVG - safe system-font fallbacks - mobile-first layout ## Brand application Use the active session brand for: - colors - typography hierarchy - logo/wordmark - buttons - borders/radii - cards/surfaces - spacing - voice Do not reset to generic gradients or neutral styling on later builds in the same session. ## Interaction quality For calculators and assessments: - inputs must be clearly labeled - values update correctly - outputs should be obvious and useful - assumptions should be visible - reset/edit behavior should work - avoid decorative complexity that does not help the user --- # Step 7 — One lightweight QA pass After creation, do **one** sanity pass only. Check: 1. the page loads (Send page, artifact, or the HTML file opened locally) 2. the main interaction/calculation works 3. there is no obviously broken desktop layout 4. there is no obviously unusable mobile layout 5. the seller brand is still present If a **critical** issue exists, fix it once in the same destination (Send's edit tool, an artifact update, or an edit to the same file) and re-check that issue. Do not spend time autonomously polishing tiny spacing, wrapping, or cosmetic details. The user becomes the next QA gate after the live link is delivered. --- # Step 8 — Return the link (or file) Show: ### Your [Lead Magnet] Is Live > [Send URL, artifact link, or file path] Then one short line: - **Send / artifact:** > It's built in **[Company]**'s brand and the same link can be edited in place. - **HTML file:** > It's built in **[Company]**'s brand. Open it in a browser, or upload it to your site or any static host to share it. Return only the one destination. Do not attach a second copy elsewhere. If the destination is not Send, do not offer to move it to Send unless the user asks. --- # Step 9 — Contextual refinement gate After the live link, offer exactly **3 context-aware edits + stop**. For an email infrastructure calculator, good examples are: - **Add another cost scenario** - **Make the assumptions editable** - **Change the design emphasis** - **Looks good — stop here** For an assessment, the options should relate to scoring/results instead. Use interactive buttons when available. Stop. ## If the user picks an edit Make the smallest useful change in the same destination: - **Send:** `GetDocument` only if needed, then `EditDocument`. Keep the same Send URL. - **Artifact:** update the same artifact, so the link stays the same. - **HTML file:** edit the same file in place. After editing, say: > Updated — the same link is live. (For a file: > Updated — same file, refreshed.) Then offer only: - **One more change** - **Stop here** --- # Fallbacks | Situation | Action | |---|---| | Website missing | Ask for website + whether a brand guide exists | | Brand guide supplied | Use it as highest authority | | No brand guide | Derive observed brand from website | | Brand ambiguous | Show concrete options and ask once | | Seller/prospect identity conflicts | Clarify roles before build | | Send unavailable | Publish as a Claude Artifact; if artifacts aren't available, deliver an HTML file | | Requested asset vague | Ask what it should calculate/assess/decide | | Research needs unknown numeric inputs | Make them editable inputs | | No logo import path | Use text wordmark; keep rest of brand | | User names a destination | Use it if available; otherwise say so and use the next one | | Critical build bug | Fix once and re-check that bug | | Minor visual imperfection | Publish; let user request edit | | User asks for another asset in same session | Reuse locked seller + brand automatically | --- # Recording rules Ideal visible sequence: 1. User: “Build me an email infrastructure calculator.” 2. If needed: website + brand-guide question. 3. `Brand I'm Using` table. 4. Short build/research activity — no long research monologue. 5. Live link (Send or artifact) or the HTML file. 6. 3 contextual refinement buttons + stop. 7. Optional one in-place edit to the same link. Do not show: - Seller Pack files - giant research reports - deal-diagnosis menus - three asset recommendations - raw HTML - the same asset in two destinations - repeated autonomous QA loops --- # Critical rules - Natural user language should invoke this skill; the user never needs the skill name. - Build the asset the user asked for. Do not recommend a different asset unless they ask for ideas. - Website missing → ask for it. - Always ask whether they have a brand guide; if not, use the website. - Keep seller identity and brand locked for the entire session. - Brand context applies to later asset requests automatically. - Publish to Send when connected; otherwise a Claude Artifact; otherwise one HTML file. - Never stop the run just because Send isn't connected. - Research only what the requested lead magnet needs. - Max 3 targeted external research lookups by default. - Once logic is clear, build. - One lightweight QA pass, not autonomous perfection. - Return the link or file quickly. - All later refinements update the same Send document, artifact, or file.