team brilliant — AI harness audit
Where does your team’s AI harness stand?
Twelve checks across the four layers. Answer honestly — it scores every layer and tells you which one to install first.
01
Knowledge
What the agent knows about your team
Your repos have agent-readable context files (CLAUDE.md / rules) capturing conventions and the why — not just code
That knowledge is committed and shared, not living on one engineer’s laptop
Someone owns keeping it current
02
Capabilities
What the agent can do, on repeat
Your team has packaged, reusable workflows — skills, plugins, or slash commands — beyond ad-hoc prompting
Those workflows are shared across the team, not stuck in one person’s setup
Someone owns the library: versioning, updates, retiring dead ones
03
Connectivity
What the agent can touch
Your agents connect to real systems — cloud, databases, browser, repos — not just chat
They connect through dedicated scoped service accounts, never engineers’ personal accounts
Those accounts are least-privilege: read-only by default, destructive actions gated
04
Governance
What the agent can’t do without permission
An allowlist lets safe, routine commands run without a human approving each one
Destructive operations are blocked or gated by hooks before they execute
You have a defined response for when an agent does something it shouldn’t
No sequence, no spam — just your result.
Answer all 12 checks to unlock your score.