Research: vercel-labs/skills ecosystem overlap with echoskill

[krisrowe]

Context

Research during Remotion skill installation revealed that Vercel operates a large-scale agent skills ecosystem that overlaps significantly with echoskill's mission.

What Vercel has

• skills on npm — owns the bare package name. CLI for installing, removing, updating, and discovering agent skills.
• vercel-labs/agent-skills — first-party skill collection. 24,312 stars, 2,212 forks.
• vercel-labs/skills — the CLI. 12,781 stars, 1,026 forks.
• skills.sh — marketplace/discovery website.
• agentskills.io — open standard for the SKILL.md format.
• 3.99 million npm downloads (Jan–Mar 2026).
• 44+ supported agents including Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Cline, Windsurf.
• Maintained by rauchg (Guillermo Rauch, Vercel CEO) — this has executive-level investment.

What Anthropic and Google say

Claude Code docs reference the agentskills.io open standard but do NOT mention vercel-labs, skills.sh, or npx skills. Gemini CLI similarly supports skills natively without referencing Vercel.

Both platforms read SKILL.md from their respective directories. The Vercel CLI knows where those directories are and writes to them, but neither platform depends on or endorses the Vercel tooling specifically.

Where echoskill overlaps

Capability	Vercel (skills)	echoskill (em)
Skill packaging format	SKILL.md (agentskills.io standard)	SKILL.md (same standard)
CLI for install/remove/list	npx skills add/remove/list	em skills install/uninstall/list
Marketplace/discovery	skills.sh website	echoskill.ai website
Multi-agent support	44+ agents via -a flag	Claude + Gemini via only/exclude
Skill repos (git-hosted)	GitHub-based, any repo	GitHub-based, marketplace registration
npm presence	skills (bare name)	aicfg / echomodel (functional names)
Stars/adoption	24k+ stars, 4M downloads/quarter	Single-digit stars, personal use

Questions to answer

• Is echoskill's value proposition differentiated from what Vercel already offers, or are we building a subset of an existing well-funded solution?
• Does echoskill offer anything the Vercel ecosystem doesn't? (e.g., echomodel's cross-platform install manifest, em skills publish, marketplace registration, platform targeting via collection conventions)
• Should echoskill pivot to being a Vercel skills provider (publishing skills to work with npx skills add) rather than maintaining a separate CLI and marketplace?
• Is there a niche echoskill serves that Vercel doesn't — e.g., smaller, curated, opinionated skill sets for specific workflows rather than a general marketplace?
• What's the relationship between agentskills.io (the standard) and Vercel (the implementation)? Is the standard genuinely open or Vercel-controlled?
• How does echoskill's em CLI compare feature-for-feature with npx skills? What does em do that npx skills doesn't, and vice versa?
• Does npx skills support user-defined skill collections — curated, hierarchical, possibly private — that let a user quickly gut-check "do I have the skills I need for context X" (e.g., home projects, work, contributing to a specific repo, authoring novels)? If not, this is the core problem echoskill set out to solve: managing your own skills across contexts, both public and private, with marketplace-like organization and discovery. This use case may be completely unaddressed by Vercel.
• Does npx skills support private skill repos or authentication for private GitHub repos?
• Can npx skills manage skills across multiple "profiles" or contexts, or is it purely per-project / per-user-global?

Clarification: these are different problems

The initial framing of this issue overstated the overlap. On reflection:

Vercel solves: "Find and install someone else's skill." It's npm for skills — discovery, delivery, installation of third-party content from a public marketplace.

echoskill/em solves: "Ensure my skills are present, current, and consistent across machines and contexts." It's closer to dotfiles-manager for skills — desired-state management of your own skill collections, syncing across workstations, knowing what you should have installed and alerting when something's missing.

These are fundamentally different categories:

	Vercel (npx skills)	echoskill (em)
Core metaphor	npm — package delivery	dotfiles-manager — desired-state sync
Primary use	Install third-party skills	Manage and sync your own skills
Answers the question	"What skills exist for X?"	"Do I have the skills I need right now?"
Cross-machine sync	No	Yes (via marketplace repos + install manifest)
Desired-state tracking	No	Yes (manifest tracks what should be installed)
Context-aware profiles	No	Potential (collections serve this role)
Public marketplace	Yes (skills.sh)	Incidental, not the goal

The "marketplace" terminology in echoskill was always more about organizing personal/org skill collections than competing with a public discovery platform. The word "marketplace" may have been misleading — "registry" or "collection" better describes the intent.

What to do

• UPDATE: Vercel IS moving into state management. Their CLI has experimental_install (restore from lock file) and experimental_sync (sync from node_modules). Plus check and update for version tracking. These are marked "experimental" but the trajectory is clear. The lock file (.skill-lock.json) already tracks installed skills with provenance. Need to evaluate how their lock file compares to em's install manifest and how far their experimental features go.
• Consider reframing echoskill language from "marketplace" to "registry" or "collection" to avoid implying competition with skills.sh
• Evaluate whether em skills install should use npx skills add under the hood for third-party delivery while keeping the manifest/sync layer on top
• Document the positioning — but be honest that it may be temporary. If Vercel completes the lifecycle (discover → evaluate → adopt into curated list → manage → subscribe to updates or snapshot/fork), the management layer stops being differentiated. The full lifecycle is: "find a skill, see how popular and trusted it is, place it in my curated collection, either subscribe to upstream updates or effectively snapshot/fork it." If their experimental_install and experimental_sync features mature into that, em's skills management is competing with a funded incumbent on their own roadmap.
• Identify what em does that Vercel cannot easily replicate: MCP server management, context unification, cross-platform config, session search. These are the durable differentiators. Skills management may not be one of them long-term.
• Acknowledge the moat risk: Even if echoskill is currently ahead on curation/management, this is a one-quarter engineering effort for Vercel if they decide to build it. They already have the lock file, the CLI, the distribution, and experimental features pointed in that direction. Skill curation is not a defensible moat against a funded incumbent already on that trajectory. The durable moat for em is the agent workstation configuration space (MCP, context, sessions) — Vercel has no business reason to build those. Skills may end up being a feature em proxies through Vercel's tooling rather than a standalone capability.
• Potential first-mover opportunity: If em integrates private collections/registries with Vercel's (and others') public platforms — so you can curate privately, graduate to public, and manage both through one tool — that's a position Vercel doesn't occupy today. They own public delivery. Nobody owns the private-to-public bridge. If em gets there first as the glue layer between private skill management and multiple public platforms, that's harder for Vercel to replicate because it requires being platform-agnostic (Vercel has no incentive to integrate with competitors' skill delivery). However: this is speculative. We're making assumptions about gaps that could be answered in an hour of research. And if the value is real, Vercel will see it too — first-mover advantage in OSS tooling is measured in months, not years.

Cross-skill synergies and centralized behaviors

• Does the agentskills.io standard define any mechanism for inter-skill relationships — shared behaviors, complementary skills, skills that enhance each other when co-installed?
• Does npx skills implement anything for this — metadata for related skills, bundles, groups, recommended pairings?
• How do skills centralize behaviors without creating hard dependencies? (Dependencies between skills is likely an antipattern — a skill should work standalone. But synergies — "this skill is better when X is also installed" — are a different thing.)
• Is there a concept of skill "profiles" or "bundles" — install a curated set for a context (e.g., "web development", "data science", "this specific repo")?
• How does echoskill's collection concept compare? Collections group skills by category but don't express relationships between them.

This is part of understanding what the standard and the ecosystem already support vs. what would need to be invented.

Related project to investigate

• ShunsukeHayashi/agent-skill-bus — "The missing runtime for Agent Skills — health monitoring, self-improvement, and dependency management for any AI agent framework." 154 stars, MIT license, created 2026-03-17. Appears to address the inter-skill dependency/management problem directly. May have Anthropic contributions — needs verification. Evaluate what it does and how it relates to both the Vercel ecosystem and echoskill.

Priority

Do the research before making any strategic decisions. Most of the questions above can be answered by reading Vercel's source code, docs, and roadmap (public GitHub issues/PRs). Don't pivot, don't reposition, don't build new features until the facts are in.