Layers
Reusable pieces of an agent. Same idea as Docker, except order only matters for conflicts and merges. Each layer caches on its own.
SlopKit Rig
Build and run agents
rig build- Compose an agent from predefined layers.
rig pull- Pull agents and layers from a repository.
rig attach- Attach into an interactive session.
curl -fsSL https://rig.slopkit.dev/install.sh | bash
agency-tester
Discover new agents
Each layer has its purpose
Each agent is composed from multiple components called "layers". You can mix them freely, re-using different parts of their setup but keeping differences where it matters.
Harness
OpenCode or Claude Code, generated from the same composed agent. Add a harness by adding a module.
Tools
You can pick any nix package, so your agents always have their favourite programs available.
System prompt
Composed AGENTS.md. Later layers override. Some paths merge.
Skills
Native. A local path or a pinned GitHub repo. Installed once. Each harness gets its own links.
Custom files
Extra files in the agent home. Later layers add or replace them.
MCP
Servers start inside the same root as the agent.
Secrets
SecretSpec scopes. Injected at process start. Not written into the image.
Sandbox
Default deny. Layers grant domains, CIDRs, write paths. No grant, no access.
Dependencies
Each layer can pull many others. Everything is composable and easily reused.
Hooks
Commands at session.start and similar. Every harness maps the same ids.
The agent does not change when the computer does.
Run locally on your machine, share with colleagues, or deploy to a remote runtime.
The agent only works on this laptop
Publish the image. Other people pull the same layers you built.
Secrets landed in the image
Layers declare SecretSpec keys. Tokens are injected at process start.
The agent can hit the whole network
Default deny. Layers grant domains, CIDRs, and write paths.
Prompt and tools drifted between machines
Composed AGENTS.md and the layer list travel with the image.
Build. Push. Attach.
And your remote rig is ready.
Build
All layers are merged into a single filesystem image and agent configuration.
rig build agency-tester
Push
You can push your built image for other people or runtimes to use.
rig push repo.slopkit.dev/agency-tester:latest
Attach
You are in the agent's environment. Each session gets its own root and sandbox.
rig attach agency-tester
Pull an agent and run it on your machine.
Pull a published agent on Getting started. Session flags, layer fields, and webhook routes are in Docs.