Source: Eliot Prince — 5 Claude Skills recipe (AI Recipe Vault, 2026-04-01) + 2026-05-02 re-fetch via Notion MCP with toggles fully expanded — full Decision Council 5-persona prompt + Daniel Priestley transcript
Author: Eliot Prince (eliotprince.com / AI Recipe Vault)
Companion video: youtube.com/watch?v=RJPzNEDYOlY
Beginners follow-up (Mar 2026): “First Steps to Using CLAUDE AI (full beginners masterclass)” — youtube.com/watch?v=wLt5COnGpe0 (46 min). Linked from the WEO course at Module 1 and Module 8 for the settings/privacy walkthrough and the voice-record-then-paste pattern for loading Claude’s Memory.
A five-skill starter set designed to make the value of Claude Skills legible in one session. Each skill is a single SKILL.md file inside its own folder — build it once, use it forever. The recipe ships four of the five as downloadable .skill attachments and walks through the fifth live. The meta-lesson is the real payload: skills compound, skills chain, and the file IS the training — there is no separate fine-tune.
Key Takeaways
- The file IS the training. A Claude Skill is a single
SKILL.mdfile. No model tuning, no agent runtime — just clear instructions Claude reads at the start of the session. - Enable the Skill Creator first (Settings → Customize → Skills) before building. It’s the meta-skill that guides Claude through structuring new skills.
- Use
/skillwhen building so Claude doesn’t veer off into general assistance mode. - Skills compound. Build one well, the next one takes half the time. By skill three the pattern is muscle memory.
- Skills chain. A “write in my voice” skill can hand output to a “branded carousel” skill. Multi-step workflows become single prompts.
- Ten minutes per skill is realistic once the Skill Creator is enabled and the source material (transcript, prompt, sample carousel) is ready.
- Review after 3 uses. The first uses surface friction points. Add rules, sharpen output format, cut whatever slows things down.
The 5 Skills
1. Decision Council — multi-persona decision maker
Spawns 5 parallel agent expert personas, runs anonymous peer review between them, and delivers a chairman-styled synthesis plus next steps. Designed for strategy calls, hiring debates, and high-stakes choices where a single-perspective answer isn’t enough. Ships as the-council.skill.
2. Framework Reverse Engineer
Methodology extractor. Feed it a link or transcript from expert content (podcast, masterclass, long-form interview) and it surfaces the hidden framework the expert is actually using — the repeatable structure underneath the examples. Ships as framework-creator2.skill.
3. Lyra Prompt Writer
Master prompt writer trained on the best prompting knowledge available online. Takes a source prompt or Reddit post and reshapes it into a production-grade, reusable prompt. Ships as lyra.skill.
4. Amazon Shopper
Personal shopper that runs via Claude in Chrome. Claude actually browses Amazon for you, asks for your preferences, then returns shortlisted options. Saves the 30–40 minutes of tab-switching a typical research session eats. Ships as amazon-shopper.skill.
5. Branded LinkedIn Carousel Generator
Trained on the user’s own best-performing content. Upload an example carousel (PDF or screenshots), the skill learns the visual + copy patterns, then outputs new carousels in that style on demand. Not shipped as a file — built live in the companion video.
Implementation
Tool/Service: Claude Cowork, Claude Chat, or Claude Code with the built-in Skill Creator enabled.
Setup:
- Open Claude Cowork (or Chat / Code).
- Settings → Customize → Skills → toggle on Skill Creator.
- Prep source material per skill: transcript or link (Framework), source prompt (Lyra), shopping preferences (Amazon Shopper), example carousel PDF or screenshots (Carousel).
- Invoke with
/skillso the Skill Creator takes over the build. - Iterate. Each skill is a single
SKILL.md; edit freely after the first 3 real uses.
Cost: Included in Claude Cowork / Claude Pro / Team / Max subscriptions — no extra fees for the Skill Creator or for installing the four downloadable skills from the recipe page.
Integration notes:
- Skills in Claude Cowork live per-account; toggle them on/off in Settings.
- For Skill 4 (Amazon Shopper), Claude in Chrome needs to be installed and granted site access.
- Multi-skill chains work implicitly — Claude picks the right skill based on context, or you can
/skillexplicitly.
Try It
- Enable the Skill Creator (Settings → Customize → Skills).
- Download the 4
.skillattachments from the recipe page and install them. - Run Decision Council on a current open question. Note what it surfaces that a single-persona chat would have missed.
- Run Framework Reverse Engineer on the best expert interview / podcast / video you’ve consumed this month. Test whether the extracted framework is sharper than your existing notes.
- Chain skills: use a voice skill to draft a post, then pass to the Branded Carousel skill to format. Time the round-trip vs doing it manually.
- After 3 real uses of any skill, open its
SKILL.mdand tighten: add one rule, sharpen the output format, cut one line creating friction.
Related
- Claude Code Skills Ecosystem — broader map of community skill libraries
- Building Skills Guide — full guide to authoring
SKILL.mdfiles - Skill Design Patterns — reusable patterns behind skills like the Decision Council
- Skills vs MCP vs Plugins — when to build a skill vs reach for an MCP server
- Skills × MCP Synergy — skills that orchestrate MCP tools
Open Questions
- Full
SKILL.mdtext for each of the four downloadable skills — not rendered inline in the Notion recipe. Would need to download the attachments and open them to capture the prompts verbatim. - Distribution model: are these
.skillfiles compatible with community skill registries (clawskills.sh, the awesome-claude-skills index), or are they Claude Cowork-specific? - How does the Decision Council’s “anonymous peer review” mechanic differ from existing community council skills (Council of the Wise, LLM Council)? Comparison would be valuable — the Notion recipe doesn’t benchmark.