SkillGild
Sign inCreate account

Overview

What SkillGild is, how the marketplace and hosted runtime fit together, and how a run works.

4 min readStart here

Choose your path

No codeI want my AI assistant to use skillsCreate an account and let your assistant connect itself. About ten minutes, no terminal skills needed.DevelopersI want to call skills from codeAPI key, one POST request, typed SDKs and signed webhooks. Five minutes to a first result.AI agentsI am an AI agentThe exact setup procedure, rules, MCP tools and what to do for every error code.CreatorsI want to publish a skillApply as a creator, write a protected prompt and take it through review.

SkillGild is a marketplace for AI agent skills and a hosted runtime that executes them. Creators publish a skill once and keep its implementation private on SkillGild's servers. Developers find skills on the website, then run them from the coding agent, terminal or application they already use.

#What is SkillGild?

SkillGild combines two products that depend on each other.

Marketplace Hosted execution
Who it serves People looking for a capability to give their agent Agents, scripts and apps that call that capability
What it does Search, categories, tags, collections, featured skills and a page for every skill Runs the skill on SkillGild's servers and returns the result
What is public Name, description, access tier, price and the input schema Only the result. The skill's private instructions never leave SkillGild.
Where you use it Browse skills, Collections Your agent over MCP, the skillgild CLI, an SDK or the REST API

Together they make a hosted skill different from a downloadable prompt file:

  • Creators keep their work private. A skill's protected prompt is encrypted at rest, decrypted only for an authorized run, and never returned by any API.
  • You get one account everywhere. A single sign-in, allowance and revocation model covers Claude Code, Codex, Cursor, Gemini CLI, any other MCP client, the CLI, the SDKs and raw REST.

#Architecture at a glance

Every decision about a run is made on SkillGild's servers: who you are, whether you have access, how much allowance you have left, and how fast you can call.

text
  Coding agent                         Your app or script
  (Claude Code, Codex, Cursor,         (TypeScript / Python / Go SDK, curl)
   Gemini CLI, any MCP client)                     │
        │ local MCP                                │ HTTPS + Bearer sg_live_…
        ▼                                          │
  skillgild CLI ── key in OS credential store ─────┤
                                                   ▼
  Browser ──▶ skillgild.dev ──────────────▶  ┌─────────────────────────────┐
              sign in with Google            │  SkillGild API  /v1         │──▶ AI model
              or GitHub                      │  catalog · runs · keys      │    provider
                                             │  creators · webhooks        │
                                             └──────────────┬──────────────┘
                                                            │
                                                 encrypted skill storage,
                                                 usage and quota records,
                                                 webhook delivery workers
Component What it does
Website Marketplace pages, your account, the creator studio, and sign-in with Google or GitHub.
SkillGild API Every /v1 endpoint: catalog, runs, API keys, devices, webhooks and the creator API.
Hosted runtime Validates input, checks access and allowance, decrypts the skill's private prompt, calls the model and returns the result.
CLI and MCP server skillgild signs you in, installs skills for your agent, runs skills, and serves them to your agent over MCP.
SDKs Small TypeScript, Python and Go clients for search, skill details and runs.
Background workers Deliver signed webhooks with retries and handle scheduled housekeeping.
Encryption Private prompts, webhook secrets and stored responses are encrypted with managed keys.

#How a run flows

  1. Your agent calls the MCP tool skillgild_run_skill. The local CLI forwards the call to POST /v1/skills/{id-or-slug}/run, using the API key from your operating system's credential store.
  2. SkillGild authenticates the key, validates the input against the skill's input schema, applies rate limits, and reserves one run from your monthly allowance.
  3. Only after those checks pass does SkillGild decrypt the skill's private prompt and call the model.
  4. SkillGild records usage, commits the reserved run, sends any webhooks, and returns the output. Your agent then applies the result with its normal file tools.

If any step fails, the reservation is released and the run does not count against your allowance. See What happens during a run for every step.

#What's available

Area Status
Catalog with search, categories, tags, collections and featured skills Available
Hosted skill runs with monthly allowances, rate limits and safe retries Available
CLI sign-in, MCP server and skill install for Claude Code, Codex, Cursor and Gemini CLI Available
API keys, connected devices and signed execution webhooks Available
Creator applications, creator studio, protected prompts and version review Available
Organizations with shared access to paid skills Available
Reporting skills and appealing decisions Available
Self-service checkout and billing Coming soon
Ratings and reviews Coming soon
Streaming and cancellable runs Coming soon
Creator pricing, payouts and analytics Coming soon
SDK packages on npm, PyPI and Go modules, and prebuilt CLI binaries Coming soon

See Upcoming features for everything we are building next.

#Where to go next