Agent-readable marketplace

Skillstore is built for agents, not only browsers.

AI agents can discover public skills, read clean Markdown context, call the Skillstore MCP server, and verify signed install contracts before installing a skill pack.

Open llms.txt View MCP metadata Browse skill packs

Machine-readable entrypoint

Use llms.txt as the first machine-readable entrypoint. It lists Markdown detail pages, the MCP endpoint, manifests, and lockfiles.

/llms.txt

Clean context pages

Use clean Markdown detail pages for concise install-focused context instead of scraping HTML.

/skills/{slug}.md /plugins/{slug}.md

MCP discovery

Use the MCP metadata endpoint to discover search, pack lookup, install, and scenario tools.

/api/mcp search_skills get_pack install_pack list_scenarios

Safe install contracts

Skill and pack manifests are Ed25519 signed, point to immutable GitHub commit artifacts, and include skillstore.lock.json payloads for verification.

How an agent should use Skillstore

/llms.txt is the starting point for canonical routes and install rules.

Use Markdown detail pages for context and JSON APIs or MCP tools for search and filtering.

Before automated installation, verify the Ed25519 signature, commit SHA artifact, and lockfile payload.

Core endpoints

GET /llms.txt POST /api/mcp

GET /skills/{slug}.md

GET /plugins/{slug}.md

GET /api/plugins/{slug}/manifest

GET /api/plugins/{slug}/lockfile

Install a pack through an agent-safe contract

MCP clients should call install_pack for a public approved pack, then verify the returned signature and lockfile before running installation.

Installation docs