meshix

shpit.dev skills

Hand-authored public Agent Skills for ShpitDev projects.

Skills are prompts, not generated docs. Each skills/<name>/SKILL.md should be small, direct, and useful to a capable model with no private project context.

Install

Install a skill with a compatible skills installer:

npx skills add shpitdev/skills --skill tabex
npx skills add shpitdev/skills --skill meshix
npx skills add shpitdev/skills --skill slant4d

For agents that read local Agent Skills directly, copy or install the folder under skills/<name>.

For local Claude Code/project validation without copying skills into your home directory, create ignored install shims:

bun run link:local

This links .claude/skills and .agents/skills to the canonical skills/ folder. bun run check validates those shims when they are present.

Agent Plugins

Repo-managed plugins live under plugins/<name> and are exposed through marketplace manifests for each harness. Prefer installing from the GitHub marketplace source for normal use.

Install Meshix for Codex:

codex plugin marketplace add shpitdev/skills
codex plugin add meshix@shpitdev-skills
codex mcp login meshix

Install Meshix for Claude Code:

claude plugin marketplace add shpitdev/skills
claude plugin install meshix@shpitdev-skills

Connect Meshix in Claude.ai or Claude Desktop:

1. Open Claude's connector settings.
2. Add a custom connector named Meshix.
3. Use https://meshix.app/mcp as the connector URL.

Shortcut: Add Meshix custom connector.

This remote connector works for direct Meshix MCP calls in Claude.ai and Claude Desktop. Current Claude connector behavior does not provide host-side polling, subscriptions, or webhook delivery, so long-running Meshix jobs must be checked with follow-up status/tool calls.

See Plugin Installation for local development, updates, duplicate cleanup, and MCP OAuth notes.

Plugin bundles are self-contained because Codex plugin archives reject skill and asset symlinks that point outside the plugin directory. The canonical skill source remains skills/<name>, and bun run check validates that bundled skill copies stay byte-for-byte in sync with the canonical source.

Meshix

Meshix turns natural-language CAD briefs into printable 3D artifacts through Studio and the Meshix MCP server. First-time agent setup starts at meshix.app/agents.

The Meshix skill is tuned for practical CAD work: choosing the right generation surface, polling long-running jobs, returning the Studio run link, and reviewing renders before a user prints anything.

Slant4D

Slant4D is a catalog-backed product for discovering, customizing, generating, and reviewing printable organizer inserts. The skill is tuned for market-derived corpus triage, upstream catalog resolution, operator MCP/API sync, source-part readiness, insert bundle readiness, and lifecycle proof planning.

Repository Layout

AGENTS.md                   # editorial guidance for maintaining skills
.agents/plugins/marketplace.json # Codex plugin marketplace
.claude-plugin/marketplace.json  # Claude Code plugin marketplace
skills/<name>/SKILL.md      # canonical hand-authored skill prompt
skills/<name>/agents/       # optional agent UI metadata
skills/<name>/assets/       # logos and other UI assets
skills/<name>/references/   # optional setup/deep detail loaded on demand
plugins/<name>/             # repo-contained plugin bundle
scripts/validate-skills.mjs # skill, marketplace, and plugin validation
scripts/link-local-skills.mjs # local ignored Claude/.agents install shims
docs/plugin-installation.md # plugin install, update, and MCP auth guide

Check

bun run check

The check validates required skill files, frontmatter shape, basic YAML-like syntax, optional agents/openai.yaml metadata, and local install shims when present. It does not generate or rewrite skills.

Design Rules

• Hand curate every skill.
• Include only what helps a new model use the public app or tool.
• Prefer live product docs, MCP metadata, or CLI --help over copied schemas.
• Remove old scaffolding when the approach changes.