marketplace

Skills for Claude Code and Kimi Code that ship code, not just suggest it — scaffold an app, get a browser-verified and AI-reviewed PR open, triage feedback, cut a branded release, round-trip client comments, and put your prompts behind a real eval gate. Sean's personal, actively-developed plugin marketplace.

_{The ship-it loop, one slash command per step. Illustrative walkthrough of real skills.}

Install

Claude Code:

/plugin marketplace add SeanningTatum/marketplace
/plugin install engineering-toolkit@sean-skills
/plugin install ai-toolkit@sean-skills
/plugin install marketing-toolkit@sean-skills

Kimi Code — everything in one shot, no clone needed:

/plugins install https://github.com/SeanningTatum/marketplace

Or per-plugin from a local clone (uses kimi-marketplace.json):

/plugins marketplace ./kimi-marketplace.json

Plugins

Plugin	What you get
engineering-toolkit	The ship-it loop, end to end — scaffold, verify, review, resolve, release.
ai-toolkit	LLM features built like code — versioned prompts, golden sets, evals that gate CI.
marketing-toolkit	The last-mile polish — make what you shipped legible and marketable.

Skills at a glance

Every skill has a README with its what / why / how and a visual of the output:

Skill	Why you'd reach for it
new-app	New SaaS app, live repo to cloned wizard, in one command instead of an afternoon of setup.
create-pr-with-review	Open PRs that are already proven to work and already reviewed — reviewers see a second draft, not a first one.
resolve-comments	Clear the easy 80% of review comments automatically; the risky 20% still needs you.
pr-format	A PR description a reviewer trusts on the first read, every time.
release	Ship notes that read like a launch, not a changelog nobody opens.
client-review	Let a non-technical client comment on your doc without a server, an account, or a login.
prompt-scaffold	Get a versioned prompt module, a hand-labeled golden set, and graders in your test runner — not a prompt string with no contract.
prompt-eval	Turn "seems better" into a pass/fail against a committed baseline, with a calibrated judge and adversarial regressions as a hard gate.
prompt-pipeline	Build the seams of a multi-step pipeline — coordinator, cheap workers, an independent checker, a repair loop that fails closed.
readme-marketing-rewrite	Rewrite a whole README in plain, marketing-grade language with real screenshots of every surface, PR'd through review.
mockup-screenshot	Give a README a visual even when there's no live demo to screenshot.

What the output looks like

Every skill ships a README with a visual of what it produces. A few of the surfaces:

_{Example outputs — create-pr-with-review, release, client-review. Each skill's README has its own.}

Develop

claude plugin validate .                          # validate the catalog (Claude Code)
claude --plugin-dir ./plugins/engineering-toolkit # test a plugin locally (Claude Code)

For Kimi Code, install a plugin locally (/plugins install ./plugins/<name>) and check /plugins info <name> for manifest diagnostics — there is no standalone Kimi validator.

See CLAUDE.md for repository structure and conventions.