Source: “I Tried 100+ Claude Code Skills. These 6 Are The Best.” YouTube tutorial, ~10-min, youtube.com/watch?v=eRS3CmvrOvA, fetched 2026-05-03. Auto-captions normalized for “Claude Code” (caption: “ClaudeCode” / “CloudCode”), “SKILL.md” (caption: “skill.md5”), “/ultrareview” (caption: “/ultra review”), “claude-mem” (caption: “ClaudeMem” / “Claude Mem”). Creator name not in transcript.
An operator-perspective curated list of six Claude Code skills/plugins (plus one bonus) chosen after the author’s claimed 400-hour Claude Code immersion across real-estate, HVAC, coaching, and marketing-agency client work. Framing: “Most people online are developing fancy skills for the sake of a cool video, but businesses don’t actually want that. They want six types of skills. They’re simple, they’re boring, but they are effective.” Distinguishes carefully between skills (single SKILL.md packages) and plugins (bigger packages containing skills + hooks + MCP servers); note that the curator’s list mixes both freely under the “skill” umbrella. Includes a sales-enablement closer: sell outcomes (10 hrs/week saved, fewer mistakes, faster leads), not workflows.
Key Takeaways
- The list intentionally excludes flashy demo-skills. Author’s thesis: real estate agents, HVAC dispatchers, coaches, and marketing agencies don’t pay for clever — they pay for time savings, error reduction, and reliability. The six picks are infrastructure for selling AI to businesses, not demo material for content creators. Aligns with the AI Marketing track outcome-first framing and the Intermediate Course’s “what does this enable for the team” lens rather than feature tour.
- Pick one and master it before scaling. Author’s onboarding advice for newcomers (line 481–485): “Don’t try to use all six of these right away. Just pick one, learn it, build out a few workflows using it, and show business owner a demo. They won’t really care about your experience, they’ll care about the value of what you’re showing them.” Same compounding-experience argument as the Cowork Jarvis tutorial’s “start today” closer.
- Skill #1 — Skill Creator (Anthropic official). The meta-skill that builds every other skill. Plain-English in → tested SKILL.md out. Removes the “I don’t know the format” wall that kills most homemade skills. Sales angle: clients don’t pay for it directly, but every billable skill comes out of it. Install:
/plugin install skill-creator. Author recommends installing globally on user scope (“no matter what project I’m working in, if I ever need to build a skill, it just automatically invokes this”). Same primitive used in the Higgsfield ad-creator workflow to package a 4-phase campaign into a one-step/ad-creatorskill. - Skill #2 — Superpowers (Obra / Jesse Vincent, ≥150,000 GitHub stars). Forces senior-developer discipline: plan first, work in isolated environments, write tests before code, brainstorm, two-stage self-review (matches-spec + code-quality). Solves the #1 Claude Code failure mode: rushed code that looks fine on the surface but breaks when the client actually runs it. Sales angle: HVAC dispatch system, marketing-agency reporting tool — the kind of project where production failure is expensive. Already documented in the Superpowers framework article; this video reinforces the install case for non-developer-led agencies.
- Skill #3 — GSD (“Get Done”) plugin. Solves context rot — the gradual quality degradation as you fill the context window in a long session. GSD spawns fresh subagents per task with clean context windows; main session stays clean; each task gets the full context budget instead of leftovers. Built-in scope-protection detection (catches when the planner silently drops a requirement) and security enforcement anchored to your threat model. Has an autonomous mode (give it a spec, walk away). Author is upfront: “GSD isn’t really a token-saving plugin. All these subagents cost tokens, but what it saves you is the hours you’d otherwise spend redoing work that Claude broke because it forgot what you asked in the first place.” Companion to Superpowers — Superpowers gives Claude the process, GSD gives it the clean context the process needs to work. Has
/gsd-helpfor command discovery. Note: distinct from the “GSD” mentioned in Simon Scrapes’ nine-component AIOS (a planning framework reference); this is a Claude Code plugin. - Skill #4 —
/reviewand/ultrareview(built into Claude Code 2.1.86+). Most operators don’t realize these are already installed./reviewruns locally, structured code review for bugs / edge cases / design issues. Fast, free beyond standard token spend./ultrareviewuploads the branch to a cloud sandbox, spins up a fleet of reviewer agents in parallel, each attacking from a different angle (logic / security / performance / edge cases). Critical detail: before any bug reaches the report it must be independently reproduced and verified — no style nitpicks, only confirmed bugs. Pricing: Pro/Max get 3 free runs; after that 20 per run depending on size. Requires Claude account auth (API key alone won’t work). Takes 10–20 min, runs in background. Use/reviewfor fast feedback on everything; reserve/ultrareviewfor pre-merge on high-stakes changes (refactors, payments, auth, DB migrations). Already documented in the Ultrareview article and the W16 release notes; this video reinforces the placement-in-workflow case (plan with Superpowers → execute with clean context via GSD → run /ultrareview before merge). - Skill #5 — Context Mode plugin. Two complementary fixes for context window pressure. (1) Garbage filtering: every tool call routes through a sandbox; raw output gets captured; only what Claude actually needs comes back to the context window. Author’s cited benchmarks: Playwright snapshot 56 KB → 299 bytes; access log 46 KB → 155 bytes; full session 315 KB → 5 KB total. Verify your own with
/contextmode:ctx-stats. (2) Event tracking: every meaningful event (file edits, task creation, decisions, errors) writes to a local SQL database. When Claude compacts the conversation, Context Mode rebuilds a session snapshot and re-injects it — files, tasks, last prompt all intact. Practical impact: sessions that used to fall apart at 30 min now run 3 hours. Install: two slash commands (specifics shown on screen —/plugin marketplace add ...-style), then restart Claude Code; auto-installs MCP server, hooks, routing. Solves the context-exhaustion failure mode more thoroughly than/compactalone. - Skill #6 — claude-mem. Cross-session knowledge persistence. Claude Code starts every session from scratch; without claude-mem you re-explain the project for the fifth time this month. Hooks into the session lifecycle, captures file edits / decisions / bug fixes / commands during the session, uses Claude’s own Agent SDK to compress them into semantic summaries, stores in a local SQLite database with vector search. Auto-generates folder-level CLAUDE.md files. Three-layer retrieval (compact-index → timeline-around-relevant → full-detail-for-handoff) reportedly delivers ~10× token savings vs dumping everything at session start. Local + web viewer for inspecting what Claude remembers. Install warning from the repo: don’t run the
npm installcommand (installs the SDK library only; hooks never register). Use the two/plugin marketplace ...commands shown on screen. Closely related to Simon Scrapes’ six-level memory taxonomy — claude-mem is the layer-3 semantic-search implementation choice (alongsidemem-search). - Bonus skill #7 — Frontend Design (Anthropic official). Install globally. Whenever Claude Code does design work, this skill suppresses the AI-generated look. Already covered in the Frontend Design deep-dive and the anti-AI-slop guide. Note: same anti-slop primitive is natively baked into Claude Design (Anthropic Labs); use the skill when you bring a Claude Design project back into Claude Code for production builds.
- Sell outcomes, not workflows. Source’s most-quoted closing argument: “Instead of offering to build an AI workflow, you’re actually just offering to save time, cut costs, and help them focus on making more profit for their business.” Maps cleanly to the Outcome Kit framing and the AI Marketing track generally. Operator implication for WEO Marketly: when proposing AI services to dental practices (the WEO Marketly client base), lead with “save the front-desk team 10 hours/week” not “we’ll build a Claude skill.”
- Skill ≠ plugin (caveats). Author calls everything a “skill” but technically: skill = single SKILL.md package (one specific job); plugin = bigger package containing multiple skills + hooks + MCP servers (changes Claude Code behavior under the hood). Of the six picks: Skill Creator is technically a plugin distribution, Superpowers is a plugin, GSD is a plugin, /review and /ultrareview are built-in commands, Context Mode is a plugin, claude-mem is a plugin, Frontend Design is a skill. The list is really “six things to install” — the marketing-friendly word choice doesn’t matter for the install commands. Reinforces the skills vs MCP vs plugins decision tree.
At-a-glance install reference
| # | Name | Type | Install | Cost |
|---|---|---|---|---|
| 1 | Skill Creator | Plugin (Anthropic official) | /plugin install skill-creator | Free; tokens only |
| 2 | Superpowers | Plugin (Obra) | /plugin install superpowers@claude-plugins-official | Free; MIT, 150k★+ |
| 3 | GSD | Plugin | /plugin marketplace add ... (shown on screen) | Free; tokens for subagent fleet |
| 4 | /review + /ultrareview | Built-in | None — Claude Code 2.1.86+ | /review free; /ultrareview 3 free runs then $5–20 |
| 5 | Context Mode | Plugin | /plugin marketplace add ... + restart | Free; runs locally |
| 6 | claude-mem | Plugin | Two /plugin marketplace ... cmds (NOT npm install) | Free; runs locally with SQLite + vector |
| 7 (bonus) | Frontend Design | Skill (Anthropic official) | Install globally | Free |
Caveat: the source video shows on-screen install commands for items 3, 5, 6 but doesn’t speak the full URLs aloud, so the marketplace coordinates aren’t transcribed. Pull the exact commands from the source video description before running.
Sales-enablement summary (source’s closer)
“I see so many people starting out trying to sell the actual workflow, but you need to be selling something else. You need to be selling the outcome.” — saving 10 hours/week, cutting admin mistakes / human error, speeding up leads, generating more leads.
- Pick one skill, master it, build a few workflows around it, demo to a business owner. They don’t care about your experience; they care about the demo’s value.
- Compounding effect: as you understand how skills compose, you build better automations faster and cheaper, which means more demos, more content, more revenue.
- Outcome-led pitch language for WEO Marketly verticals: dental practice front desk → “10 hours/week saved on appointment confirmations”; SEO client → “competitive intelligence reports 10× faster”; ad agency → “50 ad variations from one product image” (see Higgsfield ad-campaigns tutorial).
Try It
- Install Skill Creator first (
/plugin install skill-creator, user scope) before any other skill — every other skill becomes easier to author after this one is in place. - Run
/reviewon your next commit — it’s already installed and free; the answer to “should we use code review tools?” is “yes, you already have one.” - Set up
/ultrareviewwith a free run before merging the next high-stakes PR (refactor, payment code, DB migration). Compare findings vs your manual reviewer’s notes to calibrate ROI on the post-3-runs cost. - Test claude-mem on a long-running project that you’ve been re-onboarding Claude into for weeks — the cross-session payoff is most visible when there’s accumulated context to retrieve.
- Watch context overhead with
/contextmode:ctx-statsfor a real session and decide if Context Mode’s claimed 60× compression matches your usage. - Compare with the Cowork Jarvis approach — six skills for Claude Code track vs three-level rule stacking for Cowork track. Same goals (persistent memory, voice consistency, time savings); different surfaces.
- For WEO Marketly Council positioning: the six picks are largely 6-question vetting framework approved already (Skill Creator + Frontend Design = Anthropic official; Superpowers = official marketplace plugin; built-in commands need no review; GSD/Context Mode/claude-mem need standard 6-Q vetting). Treat as a starter pack rather than a custom build.
Related
- Superpowers (Obra) — full deep-dive on skill #2
- Ultrareview — full deep-dive on skill #4 (cloud half)
- Frontend Design Skill Deep Dive — bonus skill detail
- Anti-AI Slop Guide — companion to Frontend Design
- Claude Code Skills Ecosystem — broader context
- Skills vs MCP vs Plugins — the curator’s mixed-vocabulary disclaimer addressed
- Plugin Vetting Framework — for the not-yet-listed three (GSD, Context Mode, claude-mem)
- AITmpl Curated Templates Stack Builder — adjacent curation surface
- agents Marketplace — adjacent curation surface
- Marketing Skills Bundle (Corey Haines) — domain-specific alternative
- social-media-skills (Charlie Hills) — domain-specific alternative
- Simon Scrapes’ Nine-Component AIOS — claude-mem placement context
- Week 16 release notes —
/ultrareviewdebut - Claude Design (Anthropic Labs) — bonus-skill substitute when staying in Claude Design
- Outcome Kit — outcome-first framing precedent
- Cowork Jarvis Build — Cowork-track sibling
Open Questions
- Author identity not stated. Style consistent with a Claude Code creator running a free + paid community (“free school community”); 400 hours of Claude Code claimed; works with HVAC / real estate / coaching / marketing agencies. Tracking down the channel + name would help calibrate trustworthiness for WEO Council recommendation.
- Exact install URLs for GSD, Context Mode, claude-mem not transcribed. On-screen commands shown but not spoken; pull from source video description before running.
- Superpowers star count cited as 150,000+ in the source. Cross-checks against the Superpowers article (last documented at 168k stars). Likely a date-of-source artifact; both could be in-date depending on when each was authored.
- claude-mem 10× retrieval-token-savings claim is from the repo’s own README (not independently benchmarked). Worth a head-to-head comparison test against
mem-searchand against running without persistent memory. - GSD-vs-Simon-Scrapes-GSD disambiguation. Source describes GSD as a Claude Code plugin called “Get Done.” Simon Scrapes mentions GSD as a “planning framework” cited inside the nine-component AIOS. These may be the same artifact viewed at different abstraction levels (the plugin implements a planning framework) — or completely independent items sharing an acronym. Tracking the GitHub repo is the cleanest disambiguation step. Carried over open thread from Simon Scrapes ingest.