4 Agent Skills Automating Marketing Workflows

Systems Thinking Turns Marketing into Delegable Agent Skills

Builders excel at spotting repeatable processes in code but often neglect marketing due to its pull away from core work. Brian Casel reframes marketing as patterned workflows ripe for AI agents, using OpenClaw on a Mac Mini to run four custom skills via Claude. The key decision: identify manual tasks like scanning feeds or drafting emails, then encode them as skills with training data, references, and step-by-step instructions. This avoids becoming a "world-class marketer" by delegating 80% of the grunt work.

Casel rejected generic AI tools for bespoke skills tailored to his Builder Methods brand (newsletters, YouTube, Pro membership). Tradeoffs: Initial setup time (hours per skill) yields daily automation, but requires maintenance like updating RSS feeds. He stores training in markdown (voice/tone, topics) or custom apps like SparkDrop, pulling interests from years of Claude interactions. Skills reference files for sources, formats, and outputs, ensuring consistency. "We're builders. We think in systems, processes, repeatable workflows. And now that we can delegate to agents... we can apply that same systems mindset to marketing our business without needing to become worldclass marketers ourselves." (Brian Casel, introducing the philosophy—shifts builders from resistance to empowerment.)

Every role should hunt patterns: Casel urges teams to convert them into skills, scaling beyond solo use. He open-sources tools like Agent-OS and Design-OS at buildermethods.com/agent-os, providing starter kits for replication.

Radar Scan: Automated Daily Industry Intelligence Without Doomscrolling

Problem: Manually tracking AI influencers (Anthropic, OpenAI, Cursor teams) via Twitter/X wastes hours. Casel built "Content Radar Scan" to run at 4 AM, delivering a curated Markdown report by wake-up.

Process: Agent Veil fetches training from SparkDrop/markdown (voice, interests), scans RSS.app XML feeds (free/paid tiers) for 24-hour activity. Feeds track team tweets via searches like "from:anthropic OR from:whatever" converted to RSS. Agent filters for relevance (e.g., new releases, builder-relevant ideas), discards noise (promos, off-topic), summarizes into briefs with links, and outputs to dated Markdown (e.g., "2023-10-05-radar-scan.md") synced via Dropbox. Notifies via Telegram (agent channel separate from humans) with overview/link to Brainown app for reading/editing.

Why RSS.app? Handles Twitter searches scalably; XML is agent-friendly. Rejected full Twitter API for simplicity/cost. Results: Fuels content ideas for YouTube/newsletter without daily scrolling. Tradeoff: Agent judgment imperfect—requires tuning filters, but 90% hit rate per Casel. Evolution: Started with static markdown training; now dynamic via SparkDrop.

"My agent kind of collects those and analyzes a lot of like incoming feeds and then decides which ones the agent thinks based on my training are most relevant for me and my business and could potentially become material that I might use or comment on." (Brian Casel, explaining filtering—highlights training's role in personalization over raw volume.)

Brand Illustrator: Consistent Visuals from Concept to Generation

Opportunity: Public-facing assets (website, workshops, social) need unified style (coral colors, line art, shadows) without design hires. Casel created a skill in a "BM Brand Illustrations" project, blending Claude for ideation with Google Gemini API for rendering (Claude lacks native image gen).

Workflow: Manual trigger in Claude project. User dictates need (e.g., "illustration for marketing skills article: agents toolbox"). Agent interviews: context/screenshot? Colors (predefined palette)? Size/scene? Generates 3 detailed concepts (prompt-ready descriptions). User picks (e.g., Option A); agent creates dated subfolder with project.md (logs choices/prompt), generates V1 image, iterates on feedback.

References: Colors.md, idea-mapping.md, visual-world.md (object styles). Why iterative concepts? Prevents prompt drift, ensures brand fit. Tradeoff: Not fully automated (human greenlights), but 2-5 iterations yield production-ready assets matching site illustrations. Used for Pro sessions thumbnails—feels "connected" across months.

"I think that's a really important part of branding and marketing in general." (Brian Casel, on style consistency—counters one-off designs that dilute identity.)

Newsletter Pipeline: From Voice Braindump to Scheduled Send

Pain: Manual ConvertKit (kit.com) drafting/formatting for Builder Briefing (5-min weekly read at buildermethods.com) was tedious. Now, Claude skills handle 90%: content, subject, sections, export.

Two skills in "BM Newsletter" project: (1) Newsletter Writer—reviews recent examples for style, voice-dicts main message (20-min walk braindump), drafts mini-article, suggests subjects/preheaders (pick/revise), queries sections (YouTube promo? Current build?). 3-5 revisions, final hand-edit. (2) Implied exporter to Kit format (truncated, but skips UI setup).

Why voice-first? Captures raw ideas; agent polishes. Tradeoff: Still human oversight for tone, but cuts hours to minutes. Results: Consistent issues (numbered, dated, minimal HTML: mini-article, video, builds, signoff).

Video Repurposing: Sponsor-Enabled Pipeline Extension

Bottleneck: Post-recording edits for shorts (YouTube, LinkedIn). Casel spotlights WayinVideo API (wayin.ai): Upload long-form/YouTube link, AI clips engaging moments, reframes vertical, adds captions. Pay-as-you-go REST API; OpenClaw skill on Clawhub automates upload/job/pull/deliver.

Fit: Agent grabs new YouTube video, processes to clips for review/post. Why API over UI? Integrates into pipelines like radar/newsletter. Tradeoff: Cost per minute, but scales build-in-public.

"The bigger idea: every role in your org should be looking for patterns to convert into Skills that agents can own." (Brian Casel, core thesis—applies to video, extending beyond his four skills.)

"You can literally just describe what you want to do and what your business goals are. Explain those to Claude or Chatubt or Gemini and have it help you build out a custom skill that fits your specific needs for your processes." (Brian Casel, on skill-building—democratizes via natural language, no deep coding needed.)

Key Takeaways

• Scan for patterns in your marketing: List manual tasks (feeds, visuals, emails), break into steps, encode as OpenClaw/Claude skills with training refs.
• Use RSS.app for social feeds: Convert Twitter searches to XML; filter via agent instructions tuned to your niche.
• Build brand refs once: Colors/objects in markdown; Claude ideates, external API (Gemini) generates—iterate 3 concepts first.
• Voice-dictate for newsletters: Braindump → agent draft → revise; reference past issues for evolving style.
• Integrate APIs like WayinVideo: Plug into agents for post-production; start with Clawhub skills.
• Separate comms: Telegram channels for agents; Dropbox/Brainown for file handoff.
• Train incrementally: Start with markdown voice/topics; evolve to apps like SparkDrop.
• Review daily: Agents handle 80-90%, but human loop for judgment/content quality.
• Open-source starters: Fork Casel's Agent-OS/Design-OS; join Builder Methods Pro for dashboards/courses.