Subscription clients like Weglot have undocumented or partially documented brand guidelines that live in one person's head, causing inconsistency errors that require back-and-forth revision rounds and erode client trust over time. · Seen in: 2026-06-09-1359-feedback-call-weglot-x-magier-7n1I.md, 2026-06-09-0801-ticket-magier-L0xU.md

A Claude agent that accepts a Figma export or PDF of a design deliverable plus a structured brand guidelines JSON, runs a rule-by-rule compliance check, and returns a pass/fail report with specific flagged elements before the deliverable is sent to the client — wired as a required step in the delivery workflow for any client with a brand guidelines profile on file.

Build a /brand-check skill that accepts a deliverable basename and a client brand profile JSON encoding the rules Merve described: approved color combinations, text highlight usage scoped to specific sub-brands, typography hierarchy, and approved asset sources. The agent outputs a markdown checklist with pass/fail per rule and a required PM sign-off field for any fail. The Weglot brand guidelines rebuild project in Q3 is the natural moment to create the first structured profile — the output of that project becomes the input to this agent. Wire it into the Many Requests delivery workflow as a pre-completion gate for clients tagged brand-guidelines-active.

Merve named brand inconsistency as the only friction point in an otherwise strong relationship and said a clean guidelines system is the prerequisite for letting non-marketing teams commission work autonomously — building this agent during the Q3 guidelines project means the tooling ships at the same time as the guidelines, not six months later.

Designers and art directors spend repeated manual time searching Facebook ad libraries and other sources for inspiration references, with the same searches run across multiple client briefs each week. · Seen in: 2026-06-05-1146-anna-x-willi-GgAV.md, 2026-06-05-1202-ai-opportunity-mapping-ppia.md

An n8n workflow triggered by a brief submission that calls the Facebook Ad Library API and a set of curated design reference sources, filters results by industry and format type specified in the brief, and posts a ranked inspiration pack of 8 to 12 references directly into the Many Requests ticket before the designer opens it.

Build an n8n flow triggered by the Many Requests webhook on new ticket creation. The flow extracts the client industry tag and request type from the ticket fields, calls the Facebook Ad Library API with those parameters, and optionally queries a curated Airtable of past winning references magier has collected. Claude ranks the results by visual relevance to the brief description and posts a formatted Slack message to the designer with thumbnail previews and source links. The first pilot uses the Cercli and Weglot accounts where Lorenzo and Slint already have established brief formats. The Facebook Ad Library API is public and free for basic queries.

Anna and Willi both named this as a high-value time saver in the same session, and the current workaround — manually searching Magic Library and getting the same five ads — was called out by name as insufficient; every week without this tool is designer time spent on search instead of creation.

Subscription clients cancel because they believe they can produce simple assets themselves, but without brand guardrails they produce off-brand output — magier loses the account while the client produces worse work. · Seen in: 2026-06-05-1146-anna-x-willi-GgAV.md, 2026-06-05-1202-ai-opportunity-mapping-ppia.md

A Claude-powered web app where subscription clients log in, select an asset type from a menu scoped to their tier, describe what they need in plain language, and receive a brand-compliant draft built from their stored brand profile — reducing the impulse to cancel by giving clients a self-service option that stays within magier's brand guardrails.

Build a fly.io worker that serves a lightweight web UI per client, backed by a brand profile JSON stored in magier's client wiki. The client selects asset type — social post, banner, email header — describes the content, and Claude generates the asset using the brand profile as a system prompt constraint. Output is a Figma-compatible JSON or a rendered PNG. The client can request a designer review from within the same interface, which creates a Many Requests ticket. The first pilot is a subscription client currently at churn risk where the stated reason is self-sufficiency — Willi and Anna named this pattern explicitly as a retention mechanism. The brand profile JSON is the same artifact being built for the brand guidelines agent above, so the two tools share infrastructure.

Willi named this as a direct response to clients cancelling with the reason they can do it themselves — every cancellation of this type is a recoverable churn event if the tool exists, and unrecoverable if it does not.

Art directors receive vague client feedback like the design does not resonate and spend untracked time interpreting what the client means before they can act, with no learning loop across similar feedback from the same client over time. · Seen in: 2026-06-05-1202-ai-opportunity-mapping-ppia.md, 2026-06-05-1146-anna-x-willi-GgAV.md

A Claude agent embedded in the Many Requests ticket flow that reads a new client feedback comment, retrieves the last 10 feedback comments from that client as context, and posts a structured interpretation — probable intent, suggested design action, confidence level — as a reply visible only to the art director before they respond to the client.

Build a /feedback-translate skill triggered by the Many Requests webhook on new client comment in a ticket tagged design-feedback. The skill sends the comment plus the client's feedback history to Claude with a system prompt that encodes magier's design vocabulary and common feedback patterns. Claude returns a structured note: one sentence on probable intent, one sentence on suggested action, and a flag if the feedback contradicts a previous client preference. The output posts as an internal note in the ticket, not visible to the client. The client feedback history is pulled from the Many Requests API for that organization. The first pilot uses a subscription client with a long ticket history — Weglot or Cercli — where the pattern of feedback is already established enough to calibrate the model.

Willi named this as a direct time saver for art directors who currently run interpretation loops manually, and the cost of wrong interpretation is a revision round that consumes margin — the feedback translator eliminates the interpretation step before work begins, not after.

Each new ABM campaign requires a custom landing page and quiz funnel built from scratch, with messaging, quiz logic, and result copy all written manually even when the campaign structure is identical to previous ones. · Seen in: 2026-06-08-1159-abm-campaign-raci-7x54.md

A Claude agent that accepts a campaign brief — target persona, core pain point, video or ad theme, desired CTA — and generates a complete Webflow-ready landing page spec including hero copy, quiz questions with branching logic, result page copy, and a three-email nurture sequence, outputting a structured JSON that can be imported directly into the Webflow CMS.

Build a /campaign-page skill that accepts a brief markdown with five fields: persona, pain point, ad theme, CTA type, and tone. Claude generates the full page spec using the IT-frustration campaign Willi and Gabriel designed as the canonical template — hero headline, quiz question set with answer options and scoring logic, result page copy keyed to score ranges, and email sequence with subject lines and body copy. The output is a structured JSON matching the Webflow CMS schema for the magier landing page template. The Webflow developer imports the JSON and adjusts layout only. The first run uses the therapy video campaign brief from the ABM session as the seed input to validate the output quality before the campaign goes live. 1d build to wire the skill and validate against the existing campaign.

Gabriel and Willi have a live campaign with a €2,500 Webflow partnership budget and a defined launch window — building this now means the tool is validated on a real campaign rather than a hypothetical, and the output becomes the reusable template for every subsequent ABM campaign.