Practical how-to guides
Configure Clarvis, step by step.
Working files, complete examples, and a concrete check for every important part of the agent stack.
Start with an agent and its specialists, add reusable knowledge, then set the runtime boundary before you package and share extensions. Every guide begins with the smallest useful configuration and grows from there.
Inside every guide
Enough detail to build it yourself.
-
01
Exact path
Know which file to create.
-
02
Minimum config
Begin with a complete working shape.
-
03
Real patterns
Open full examples for common work.
-
04
Verification
Check both capability and boundary.
Build the agent team
Create a focused role, connect bounded specialists, and give the team knowledge it can load only when relevant.
Configure an agent and sub-agents
Create a lead, define which specialists it may spawn, and give every role its own model and authority.
Open guide ↗Create a reusable skill
Write SKILL.md, bundle an optional reference, and make the method discoverable without filling every run context.
Open guide ↗Set the runtime boundary
Control which tools exist, which actions may proceed, where commands run, and what must be checked every time.
Configure grants and guard
Separate per-agent tools from workspace command policy, with working allow and deny examples.
Open guide ↗Configure the sandbox
Contain command execution with an explicit filesystem, network, and fail-closed availability posture.
Open guide ↗Write a lifecycle hook
Block a dangerous call before it runs or refuse completion while the test suite is failing.
Open guide ↗Package and distribute
Turn a working capability into a reviewable plugin, then place supported plugins in a catalog your team controls.
Connect systems you own
Preserve the agent-facing contracts while knowledge and plans live in the systems your organization already trusts.
Understand the stack
Build the first agent team.
Create one lead and two specialists, then verify that each role receives only the authority it needs.
Start with agents