Skills

Skills for AI coding agents, installable with one command.

Installation

One command installs this catalog and the four upstream skills through the skills CLI:

npx -y github:Vivswan/skills -g                   # one copy under ~/.agents/skills, for the agents that read it
npx -y github:Vivswan/skills -g -a claude-code    # the same copy, plus a symlink to it for Claude Code

Each npx skills add it runs is unattended (-y), so nothing prompts and an installed copy is overwritten. Its flags pass through to every call (-g for your user, -a <agents> for the agents, --skill <name> for one skill, which runs only the call that ships it), and --help prints the five commands without running them.

Every call installs to universal: the one ~/.agents/skills directory that Codex, GitHub Copilot, OpenCode, and most other agents read. Claude Code, Junie, and Cursor read their own directories, so -a names them to get a symlink each. Without -a universal behind the scenes, the CLI would install to every agent it knows.

A call that fails does not stop the others. The run ends with the failed calls listed, each with how it ended and the lines the CLI said, and exits 1:

install: 1 of 5 skills add calls failed
  Leonxlnx/taste-skill/skills/taste-skill: exit 1
    Skipped /tmp/skills-ME8i1r/skills/taste-skill/SKILL.md - YAML parse error: Nested mappings are not allowed in compact mappings at line 2, column 14
    No valid skills found. Skills require a SKILL.md with name and description.

Or as a Claude Code plugin. The root .claude-plugin/marketplace.json publishes this catalog as the vivswan-skills plugin, and each upstream skill by its own name (/plugin install unslop@vivswan-skills):

/plugin marketplace add Vivswan/skills
/plugin install vivswan-skills@vivswan-skills

Skills come first on Claude Code, Codex, and GitHub Copilot alike: each SKILL.md is the whole workflow, and the plugin manifests are optional (the rules).

Available Skills

Automatic

The agent applies these on its own when the task matches:

• /code-standards - Maintainability standards for code changes, folded into reviews
• /codex-browser-fix - Codex only: keeps the bundled Chrome plugin driving any Chromium browser (Chrome, Edge, Brave, Vivaldi, Opera) through app updates and custom model providers
• /craft-skills-and-memories - Create and repair skills and memories at their canonical source
• /never-twice - Climb every repeated failure to the most durable fix reachable
• /no-invalid-states - Invariants in the type system instead of runtime checks
• /pr-and-issue-discipline - PRs and issues that show the change instead of describing it
• /project-naming - A name for a project, package, or paper from one table of checked, scored candidates
• /pr-landing-discipline - Draft flips, review rounds to convergence, line accounting, and a human-gated merge
• /reply-and-review-discipline - Plain-first replies to issue reporters and one pending review on someone else's PR
• /rubber-duck-review - Cross-model, read-only second-opinion code review
• /session-events - One daemon per session delivers events from any polled source as messages in the peer envelope, quoting nothing from the polled system; the daemon behind /watch-github

Invoked by you

Load only when you invoke them (/skill-name in Claude Code, $skill-name in Codex):

• /claude-remote-peers - Message a Claude Code session on another machine over ssh: two socket forwards, no copies, pushed replies
• /natural-writing - Prose without AI writing tells
• /orchestrator-mode - Parallel worktree subagents with gated landings, direct or PR-based

Upstream skills

Skills written in other repositories. Nothing of them is copied here: the marketplace fetches each from its upstream's default branch, so an install is always upstream's latest, with upstream's own frontmatter.

• /unslop - Cut AI tells from any writing (from cursor/plugins)
• /frontend-design - Distinctive visual design for new or reshaped UI (from anthropics/skills)
• /design-taste-frontend - Anti-slop frontend for landing pages, portfolios, and redesigns (from Leonxlnx/taste-skill)
• /impeccable - Design direction for frontend work: design, critique, audit, polish, or harden an interface, with detector rules against AI-template tells (from pbakaus/impeccable)

Installed by npx -y github:Vivswan/skills -g or bun run add [-a <agents>], frontend-design, design-taste-frontend, and impeccable load only when you invoke them: the install sets disable-model-invocation: true and policy.allow_implicit_invocation: false on the copy, and an npx skills update that refreshes the copy undoes it until you install again. The direct lines below and a Claude Code plugin install keep upstream's own trigger; unslop ships its own disable-model-invocation: true.

Each one alone, through the skills CLI:

npx skills add cursor/plugins/pstack/skills/unslop -g
npx skills add anthropics/skills/skills/frontend-design -g
npx skills add Leonxlnx/taste-skill/skills/taste-skill -g
npx skills add pbakaus/impeccable/.agents/skills/impeccable -g

How the skills reference each other (an arrow means "mentions and hands off to, where installed"):

---
config:
  layout: elk
  elk:
    nodePlacementStrategy: NETWORK_SIMPLEX
---
graph LR
  wt["/working-text"]
  cs["/code-standards"]
  nw["/natural-writing"]
  om["/orchestrator-mode"]
  pid["/pr-and-issue-discipline"]
  rrd["/reply-and-review-discipline"]
  rdr["/rubber-duck-review"]
  pld["/pr-landing-discipline"]
  pnm["/project-naming"]
  nis["/no-invalid-states"]
  nt["/never-twice"]
  vwc["/verify-with-controls"]
  csm["/craft-skills-and-memories"]
  wgh["/watch-github"]
  se["/session-events"]
  wh["/worktree-hygiene"]
  us["/unslop"]
  cbf["/codex-browser-fix"]
  crp["/claude-remote-peers"]
  fd["/frontend-design"]
  dtf["/design-taste-frontend"]
  imp["/impeccable"]

  cs <--> wt
  cs <--> rdr
  csm <--> nt
  nis <--> rdr
  nt <--> rdr
  pid <--> pld
  pid <--> rrd
  pld <--> wgh
  rdr <--> rrd
  rdr <--> vwc

  wt --> rdr
  wt --> pid
  nw --> wt
  om --> rdr
  om --> wgh
  om --> se
  wgh --> se
  om --> cs
  om --> pid
  om --> pld
  om --> wh
  rrd --> wt
  pld --> rdr
  cs --> nt
  cs --> nis
  nt --> nis
  om --> vwc
  pnm --> vwc
  pnm --> om
  csm --> nw
  csm --> wt
  om --> wt
  wt --> us
  pid --> us
  rrd --> us

  click wt "./skills/working-text/SKILL.md"
  click cs "./skills/code-standards/SKILL.md"
  click nw "./skills/natural-writing/SKILL.md"
  click om "./skills/orchestrator-mode/SKILL.md"
  click pid "./skills/pr-and-issue-discipline/SKILL.md"
  click rrd "./skills/reply-and-review-discipline/SKILL.md"
  click rdr "./skills/rubber-duck-review/SKILL.md"
  click pld "./skills/pr-landing-discipline/SKILL.md"
  click pnm "./skills/project-naming/SKILL.md"
  click nis "./skills/no-invalid-states/SKILL.md"
  click nt "./skills/never-twice/SKILL.md"
  click vwc "./skills/verify-with-controls/SKILL.md"
  click csm "./skills/craft-skills-and-memories/SKILL.md"
  click wgh "./skills/watch-github/SKILL.md"
  click se "./skills/session-events/SKILL.md"
  click wh "./skills/worktree-hygiene/SKILL.md"
  click us "https://github.com/cursor/plugins/tree/HEAD/pstack/skills/unslop"
  click cbf "./skills/codex-browser-fix/SKILL.md"
  click crp "./skills/claude-remote-peers/SKILL.md"
  click fd "https://github.com/anthropics/skills/tree/HEAD/skills/frontend-design"
  click dtf "https://github.com/Leonxlnx/taste-skill/tree/HEAD/skills/taste-skill"
  click imp "https://github.com/pbakaus/impeccable/tree/HEAD/.agents/skills/impeccable"

License

Individual and Small Organization License; the version and terms are in LICENSE.md.