Your coding agent is fast, capable, and completely unsupervised.
This pack is the supervision. 27 open-source skills that read every diff like a senior reviewer, hand down a blunt A-F ship grade, and block the merge until it holds. Then they handle the parts of shipping nobody prompts for: design, copy, SEO, and Swift/iOS and Android. One of them argued Amazon out of $448.31 in one sitting. Your pull requests don't stand a chance.
/plugin marketplace add JasonColapietro/suede-creator-skills
/plugin install suede-skills@suede
Every skill is a plain-Markdown SKILL.md folder you can read before you enable it. No uploads, no registry writes. More install paths for Codex, Cursor, and 25+ agents.
The only pack that beat both GSD and Superpowers — on the same rubric, losses published
Self-graded on the same 15-category ship-gate rubric, run on all three packs. Losses published too: see the full scorecard.
Changelog · straight from main
A review pack had better survive its own review.
So the work ships in public. Every entry below is a real commit on main: what changed, when it landed, and the hash that proves it. No "various improvements."
Last shipped Jul 19, 2026
161 commits · 10 weeks · newest on the right
-
Android lane opens: android-app-factory + suede-recommend-next-action
One prompt to a Play-Store-ready Kotlin/Compose build, plus a skill that reads repo state and names the single next move. The pack goes 25 → 27.
-
MCP server finds its skills from any install path
suede_creator_mcpnow resolves its catalog throughCLAUDE_PLUGIN_ROOTinstead of assuming a fixed layout, so marketplace-cache installs stop coming up empty. -
Homepage v2: the hero terminal rotates all six lanes
Ten commits in one sitting: a live sample session that cycles disciplines, a lane index, and copy with an actual hook.
-
Consumer-recovery lane: the negotiator points at your bills
amazon-returns-recovery runs fee discovery and live-chat disputes; subscription-recovery generalizes it to any recurring charge.
-
LCP: above-the-fold previews load first
Skill preview cards above the fold get
fetchpriority="high"with async decoding; everything below the fold stays lazy. -
Artist profile no longer returns zero skills
The MCP artist filter matched a
skill.areano catalog entry has; it now maps to the creator-tagged skills.classify_filealso stops reading ancestor directory names outside the project. -
site-to-ios-app: live URL to App Store archive
Wrap an existing site into a submittable iOS shell. The mobile lane opens.
-
Skill-count drift guard in the validator
The build now fails whenever public copy disagrees with the number of skills actually in the pack. This page can't quietly lie.
One pack, six disciplines
Code review is just the opening act.
The same discipline that grades your diffs also designs the page, writes the copy, ships the iOS and Android app, packages the release rights, and argues your money back. Six lanes, 27 skills, one install.
Benchmarks: self-graded, losses shown
Suede against GSD and Superpowers, losses included.
We ran our own 15-category ship-gate rubric on Suede Creator Skills, GSD, and Superpowers the same way. The full category breakdown is below, including the rows where GSD or Superpowers beats Suede.
| Use case | Winner | Grade |
|---|---|---|
| Suede product / company work | Suede | A+ |
| Public copy, design, SEO, rights, launch | Suede | A+ |
| Long project lifecycle | GSD | A+ |
| Strict engineering discipline | Superpowers | A+ |
| Multi-agent orchestration | Suede / GSDtie | A |
| Skill ecosystem maintainability | Superpowers | A+ |
| Ship-risk grading | Suede | A+ |
Suede wins outright on 3 of 7 use cases, plus a tie with GSD on multi-agent orchestration. GSD takes long project lifecycles. Superpowers takes strict engineering discipline and skill ecosystem maintainability. Install the pack that matches this week's work.
See the full 15-category scorecard
| Category | Suede Skills | GSD | Superpowers |
|---|---|---|---|
| Overall | A- | A- | A |
| Breadth / Coverage | A+ | A | B |
| Planning / Specs | A- | A+ | A |
| Execution Workflow | A- | A | A |
| Debugging Method | A- | A | A+ |
| Testing Discipline | B+ | A | A+ |
| Verification Discipline | A | A | A+ |
| Git / Worktree Hygiene | B+ | A | A+ |
| Durable State / Resume | A- | A+ | B+ |
| Code Review / Ship Gate | A+ | A- | A |
| Skill Authoring / Governance | B+ | B+ | A+ |
| Evidence / Legal Boundaries | A+ | B | B |
| Domain Expertise | A+ | B | C |
| Install / Portability | A- | B+ | A |
| Estate Cleanliness | B+ | A- | A |
Grades come from our own ship-gate rubric, run against all three packs with the same categories. Not an independent audit. We're publishing the rows where the field beats us on purpose, including the ones that say GSD or Superpowers outright win.
What your agent learns.
Claude judges.
Codex generates.
Claude decomposes a high-volume job, writes self-contained briefs, spawns parallel OpenAI Codex CLI workers billed to your ChatGPT plan, then reviews every output against acceptance criteria before anything ships.
Review that ends in
a blunt A-F grade.
Deep findings across TypeScript, React, Next.js, Swift/iOS, OWASP, and database checklists, then a letter verdict across 7 evidence-backed lanes. A hardcoded secret is an instant F with the exact file and line — no other lane can raise it. iOS gets its own traps list: force unwraps, main-thread violations, and native contract drift that crashes shipped apps.
Every capability is an inspectable SKILL.md folder the agent loads on demand, so context stays lean and the job stays in focus
Runs scout, build, review, verifier, and release lanes that share context instead of acting like siloed subagents — with WIP collision detection and a signed handoff
Turns LLM, RAG, classifier, and agent surfaces into measurable AI-SPECs with failure-mode rubrics, eval cases, and acceptance gates you can fail a release on
Writes CI that actually gates the merge: stack and lockfile detection, one required check that can't deadlock, and branch protection that holds
Point it at any URL. It reverse-engineers the layout logic, type rhythm, and color strategy that make that site work, then rebuilds your page in that language — your own copy, your own assets, nothing traced or copied
Runs the same persuasion frameworks agencies bill five figures to teach — AIDA, PAS, StoryBrand, JTBD — plus 12 headline formulas and a 29-point anti-slop gate before a word ships
Nine audit lanes in one pass — technical, intent, metadata, structure, schema, AI citation readiness, copy quality, E-E-A-T, topic clusters — wider than any single-purpose SEO skill in the wild
Scores brand and page visibility A-F across first-screen clarity, CTA pull, proof, AI readability, schema depth, and search result fit
Takes a live URL through the App Store 4.2 wrapper-risk gate, a native or Capacitor shell, and screenshots polished enough for a downtown agency pitch deck, not a template generator
Commits to a named aesthetic lane, rejects the first reflex and the second-order trap, and builds color, type, and motion up from that decision — never a template with your logo dropped in
The same negotiation discipline pointed at your own accounts: one skill argued Amazon into refunding $448.31, the other audits Netflix, gyms, and every App Store subscription you forgot you had
Say "Cue Suede" mid-run and choose: change something, preserve what worked, or accept as-is. The skill mimics what worked on every pass after — across design, copy, agent-team, and launch-packaging lanes
One prompt to a Play Store-ready native app: Kotlin + Jetpack Compose scaffolding, Play Billing, ASO listing copy sized to Play's character limits, and a signed release gated the same way as everything else in this pack
Ask "what's next" and get one recommended move — scored on goal fit, urgency, and leverage — packaged as a short prompt you paste and run, expandable into a full operator brief only when you ask
The router
Don't memorize 27 commands.
Ask in plain words. The umbrella skill, /suede-workflow-skills, reads the request and loads the right lane — one install, every skill reachable. Agents that speak MCP get the same routing as structured tools: list_suede_skills finds the skill, get_suede_skill loads it.
The optional MCP server ships in the repo, runs local with zero dependencies, and adds install options, audit scaffolds, grading, and QA checklists on top of discovery. Wire-up lives on the install page.
Install the open-source pack.
Two lines in Claude Code install the full 27-skill pack. Prefer less? suede-agent-workflows@suede and suede-code@suede install focused subsets, and install.sh covers the clone route. Either way the agent gets orchestration, code review with A-F grading, AI evals, design, copy, SEO, and CI rails it can reuse across your own work. Bring your own company, repo, page, or product.
/plugin marketplace add JasonColapietro/suede-creator-skills
/plugin install suede-skills@suede
npx skills add JasonColapietro/suede-creator-skills
git clone https://github.com/JasonColapietro/suede-creator-skills.git && bash suede-creator-skills/install.sh
If the pack saves you an hour, star the repo — stars are how other builders find it.
The full pack
All 27 skills.
Every skill is a plain-Markdown SKILL.md folder you can read before you install it. Click any row for its docs, or browse the full catalog with install paths.
Workflow evidence
Not prompt soup.
A portable operating system for agents.
The repo ships inspectable skills, public docs, install paths, MCP source, copy banks, safety boundaries, linked preferences, and feedback loops that follow the work. The GitHub Pages site itself proves the bar: public docs that look, read, link, and grade like a real product surface.
Operator feedback
agent · before handoff
- Command the agent mid-run when something hits or misses, then make the next pass sharper.
- The skill links to preferences and context instead of stuffing everything into one swollen page.
- Taste memory. Say what worked and the skill preserves it, then mimics that pattern on every later pass instead of guessing fresh each time.
For musicians and labels
Releasing your single
gets the same QA spine.
Campaign planning, sync packaging, release linting, and rights evidence live in the same pack as the code review and design lanes.
- suede-campaign-in-a-box turns one release into a full rollout: eras, hooks, stunts, and fan missions.
- suede-rights-passport and suede-rights-audit turn provenance, splits, and ownership gaps into evidence a label or supervisor can actually review.
- suede-release-linter and suede-sync-packaging catch broken metadata and build the supervisor-safe one-sheet before the drop goes out.
Two commands from here
Stop shipping on vibes.
Give your agent review with a spine, CI that holds the door, and copy that earns the click. MIT licensed, plain Markdown, readable before you run it.
/plugin install suede-skills@suede
Questions, answered.
What are Agent Skills?
Agent Skills are SKILL.md folders that Claude Code and OpenAI Codex load on demand to follow a documented workflow. This pack ships 27 of them under the MIT license: orchestration, code review and grading for web stacks and Swift/iOS, a site-to-iOS app conversion workflow, a one-prompt Android app factory, CI ship-gates, AI evals, design, copy, SEO, and a creator toolkit for music rights and release prep. Every folder is plain Markdown you can read before you install it.
How do I install the pack?
In Claude Code, run /plugin marketplace add JasonColapietro/suede-creator-skills, then /plugin install suede-skills@suede. Prefer a clone? install.sh copies all 27 skills into ~/.claude/skills/. Codex installs any skill folder with its built-in skill installer. Full paths live on the install page.
Does it work with both Claude Code and Codex?
Yes. Each skill is a standard SKILL.md folder, the format both runtimes read. Claude Code installs via the plugin marketplace or a folder copy into .claude/skills/; Codex installs via its skill installer into ~/.codex/skills.
Is it safe to run?
The skills inspect local files and pages and write local reports. They do not upload files by default, write to any registry, request secrets, or claim legal clearance. The whole pack is MIT licensed and inspectable before you enable anything.