Jonathon's AI Wiki

Seven Claude Skills That Actually Run My Business

9 min read

Source: raw/I_Tried_100+_Claude_Skills._These_7_Actually_Run_My_Business.md (YouTube transcript, 8piX0z_HI9I, fetched 2026-05-15). Creator runs a paid AI-playbook newsletter and Q&A community; not named in the transcript.

A curated list of seven skills that survived the “tested 100+, kept 7” filter for an AI-content business operator. The skills are explicitly framed as folders with instructions — the creator’s load-bearing thesis is that skills are platform-agnostic and durable across model and tool generations because “the pattern underneath has been the same since the 1990s.” Companion to Nine Claude Code Plugins to Build 10× Faster and Six Best Claude Code Skills for Business — three independent curated lists from the same week.

Key Takeaways

  • Anthropic’s skill-creator skill is the entry point. Describe what you want in plain language; it interviews you, then writes the SKILL.md and any reference files. Two modes: create a new skill (interviews + drafts + does any background research the skill might need) or improve an existing skill (reads your current SKILL.md and rewrites against Anthropic’s best-practice guidance). The “if you take one thing away” framing in the source: you don’t need to be technical to build Claude skills.
  • /today daily digest skill replaces the 20–25-minute every-morning ritual of Calendar → Gmail → Slack → task-list manual triage. The skill knows which calendars to pull from, which Slack channels matter, which team-member emails to prioritise. Output renders both in Claude Code and into Obsidian as a daily-plan page. The “structure works for any business — swap in your sources” framing makes this a template, not a one-off.
  • /watch video skill (credit: Brad at Brad Automates — bradautomates/claude-video, see skill-bundle-build-pattern) gives Claude visibility into Loom and YouTube videos by pulling the transcript plus extracting key frames. Without leaving Claude Code, you can ask questions about what’s in any video — competitor analysis, replay calls, reference content for later. Composable into other skills (see article-optimizer below).
  • /ai-digest skill reads incoming AI newsletters from Gmail and posts a Slack summary with headlines, descriptions, key insights — and is customizable in plain language (“I want to update the AI digest skill to remove it pulling new tools and productivity”). The creator emphasises the rapid update loop: skill editing is just a conversation, not a code change. Customisation note: same structure works for industry reports (real estate), market data (ecommerce), or any inbox-filtering workflow.
  • /newsletter skill turns a 90-minute Q&A-call transcript into newsletter topic ideas. Uses XML-tagged transcript input or pulls from a transcript database (the creator originally used Supabase; migrating to Notion). Output is three ranked topics with rationale + recommendation; the creator picks one, the skill drafts the section against it. Works for any recurring meeting recording — coaching calls, onboarding calls, support calls.
  • /article optimizer skill converts any piece of content (YouTube transcript via the watch skill, blog post, podcast transcript) into a long-form article structured for AEO/GEO — Answer Engine Optimization and Generative Engine Optimization — so it gets cited by ChatGPT, Perplexity, Google AI Overview. Output written in the user’s voice, not generic AI voice, with quick-answer block + key takeaways + full article. Auto-saved to Obsidian per skill instructions. The creator’s framing: “you’re sitting on a ton of material that can be repurposed into articles AI search engines can find and cite.” See FLUQs framework for the underlying AEO/GEO pattern.
  • /decide decision-memo skill is the skill the creator highlights as the most architecturally interesting. Multi-agent pipeline: (1) interview agent asks one question at a time — options, why now, who’s affected, gut leaning; (2) structure-options agent lays out explicit assumptions + pros/cons; (3) contrarian agent looks for hidden assumptions, second-order effects, places where the emotionally easier option is masquerading as the strategically right one; (4) weigh-and-recommend agent returns the verdict including reversible-vs-irreversible classification + immediate next action. Output saved as a decision memo in Obsidian. Customisable contrarian-pass focus + addable mental models.
  • Composition is the multiplier. The article-optimizer skill explicitly invokes the watch-video skill. The newsletter skill writes into the article-optimizer pipeline. The decide skill writes into the same Obsidian vault as the daily-digest skill. Skills-calling-skills is the unlock — see Skill Design Patterns for the structural pattern.

The “folders with instructions” framing

The creator’s closing claim is the most durable takeaway:

These are just folders with instructions. That’s all they are. And I know that sounds simple, but that’s actually the point because everything in AI is going to keep changing. The tools will change, the models will change. But this concept that we’ve been talking about here today, writing instructions that an AI can follow and reuse, that’s been the pattern since the 90s and before, and it’s not going anywhere.

This matches the W21 harness-portability framing from Sigrid Jin (harness + model are swap-replaceable, context is the asset that compounds) and the Angela & Caitlin Agent Platform team interview — Anthropic’s “harness + model are pairing, not commoditising” position from the platform team. The creator’s operator-side conclusion (skills survive tool churn) lines up with Anthropic’s platform-side conclusion (skills are deliberate opinionated primitives).

Multi-agent pipeline pattern — the /decide skill

The decision skill is the most replicable architecture in the catalog. Pipeline:

  1. Interview agent — single-question-at-a-time elicitation. Forces the user to articulate options, motivation, affected stakeholders, current gut-leaning.
  2. Structure-options agent — formal options table with explicit assumptions + pros + cons.
  3. Contrarian agent — adversarial pass. Hidden assumptions, second-order effects, emotional-easy vs strategically-right divergence.
  4. Weigh-and-recommend agent — final synthesis with verdict + reversible/irreversible classification + immediate next action.

Cited example: a $hire-a-growth-coach? decision returned “don’t hire a coach right now; revisit in 90 days; your revenue gap looks like a focus gap, not a strategy gap.” The contrarian pass surfaced execution-backlog framing the user hadn’t named explicitly.

Same multi-agent pattern as Managed Agents advisor-strategy (one orchestrator, multiple specialists) and Agent Teams / Subagents dispatch — but compressed into a single SKILL.md instead of a separate Managed Agents deployment. Suggests the skill-as-multi-agent-pipeline pattern is a viable lightweight alternative to CMA when the pipeline is small (≤4 sub-agents) and doesn’t need persistence between invocations. ^[inferred]

Obsidian as the persistence layer

Four of the seven skills (daily-digest, article-optimizer, decide, implicitly newsletter) write outputs into Obsidian. The creator’s pattern:

  • Skill produces output → writes markdown to a specific Obsidian folder
  • User can open Obsidian → read/edit/expand
  • The Obsidian vault accumulates a personal knowledge base of decision memos, daily plans, AEO-optimized articles, newsletter drafts

This is the operator-side equivalent of the Karpathy LLM-wiki pattern — instead of an LLM-maintained wiki, it’s an LLM-maintained personal knowledge base, with Claude as the librarian writing into Obsidian. ^[inferred]

The decision memo path specifically is worth lifting as a standalone reusable artifact: SKILL.md that interviews → multi-agent passes → writes a dated decision memo at decisions/YYYY-MM-DD-decision-name.md with the full report + verdict + next action.

Try It

  1. Install Anthropic’s skill-creator first. Whether in Claude Code or Cowork (/skill), this is the entry point. Use it to build your first skill — pick a 20-minute daily task as the target.
  2. Build a /today daily-digest equivalent matched to your stack — swap the calendar/email/slack triple for whatever tools you check every morning (CRM dashboard, project-management tool, market data). The structural pattern is portable.
  3. Install Brad Automates’ watch-video skill (bradautomates/claude-video). Useful immediately for competitor analysis, internal call review, and as a building block for the article-optimizer pattern. See skill-bundle build pattern for the dual Claude Code + Cowork distribution.
  4. Build an inbox-triage skill (the /ai-digest pattern) for whatever recurring information stream you currently scroll-through manually. Industry newsletters, vendor changelogs, GitHub-star feeds — same shape.
  5. Build a transcript-to-content skill (the /newsletter pattern). If you record any recurring call (coaching, support, onboarding, sales), this skill is the direct pipeline from those recordings to content ideas.
  6. Build an AEO/GEO article-optimizer skill that calls the watch skill as a sub-skill. Convert one piece of existing content (a YouTube video, podcast, blog post) into an AEO/GEO-optimized article. Confirm AI-search engines start citing it within 30–60 days.
  7. Build a /decide decision-memo skill with the four-pass pipeline (interview → structure → contrarian → weigh-and-recommend). Customise the contrarian-pass focus to your business — for an operator: focus on focus-vs-strategy gap, capital-allocation, reversibility. For a research role: focus on confounders, sample-size, alternative explanations.
  8. Compose skills into pipelines. The watch → article-optimizer chain is the simplest composition; build one yourself and you’ll see how skills-calling-skills changes the unit of work from “one task” to “one workflow.”
  9. Pair this catalog with Nine Claude Code Plugins to Build 10× Faster and Six Best Claude Code Skills for Business — three independent curated lists, useful as a triangulation check.

Open Questions

  • The creator runs a paid “AI playbook” community + weekly Q&A calls, but is not named in the transcript. Worth a research pass to identify — the named-creator helps with downstream credibility weighting and for finding the actual SKILL.md sources (the video promises links in the description but those aren’t captured in raw/).
  • All seven skills are referenced as “downloadable in the description” — the actual repo URLs aren’t captured in the source. Without the SKILL.md sources, the architectural claims (multi-agent decision pipeline, watch-video sub-skill composition, AEO/GEO article structure) can’t be independently verified against the implementation. Flag for follow-up.
  • /decide skill’s “multi-agent passes” claim should be cross-checked against the actual SKILL.md — if it’s literally one SKILL.md with sequential prompts, the multi-agent framing is rhetorical; if it spawns sub-agents via Task / Agent tools, it’s structurally multi-agent. The distinction matters for the skill-design-patterns taxonomy.
  • The Brad Automates watch-video skill is mentioned with attribution but no link in the source. Memory indicates bradautomates/claude-video is the repo; worth confirming and possibly drafting a standalone wiki article on the watch skill since multiple curated lists now reference it.

Graph View

Backlinks