2.4 Skills, MCP & Prompts
Every skill, MCP server and instruction file of every agent on this machine, in one table. Three tabs, one per kind.
The reason the three share a page: they are the same question asked three ways — what does this agent carry, and which of the others carry it too. A skill installed in one agent and missing from another shows up as a gap in the row.
Skills
Install a skill into one agent or several at once, from:
- a GitHub repository,
- a
.zipor.tar.gzarchive, - or a folder on this machine.
The table lists which agents have it. Open one to read it, copy it into another agent, or delete it.
MCP servers
Add an MCP server the same way — as the JSON block a server is written in, which is the format every vendor's "add this server" snippet already comes in.
Ticking more agents writes the same server into each one's own configuration file, in that agent's format.
For servers that need a credential — GitHub, Notion, Lark — prefer Connections, which keeps the token in Gateway instead of copying it into every agent's configuration file.
Prompts
The instruction file an agent reads before every session — CLAUDE.md, AGENTS.md and their equivalents. Edit it here, open it, delete it, or copy one agent's into another.
Copying between agents
Every row is per item, with a column per agent, so the whole page answers one question: what does this agent have that that one does not. Copying is picking the agents to add it to.
Which agents can take which of the three is in the MCP · Skills · Prompt columns of the table in the README.
Next: 2.5 Connections.
2.3 Monitoring, sessions and activity
Three different records, from three different sources: the transcripts agents write anyway, the monitoring hook Gateway installs, and the network connections of the agents' own processes.
2.5 Connections
A connection is an MCP server with its credential kept in Gateway rather than in every agent's file.