Sections
On this page
What a Harness is
A Harness defines a repeatable way for agents and people to investigate a question, record a decision, and turn it into the next plan of work.
A Harness is neither a chat nor a prompt collection. It is a versioned process: roles, a step graph, access rules, limits, expected inputs, and outputs. The same Harness can be safely repeated for different work.
- 01Version
- 02Run
- 03Analysis
- 04Review
- 05Decision
- 06Approval
- 07Task plan
When to use it
Architecture choices
Independent experts propose options, test assumptions, and leave an explainable decision.
Risk and security
Dedicated roles look for threats and objections; risky actions pass through a human confirmation.
Decomposition
After a decision, the system prepares an Flownix task draft for a person to review before applying it.
The core principle
Instead of retaining hidden reasoning, a Harness records inspectable work artifacts: proposals, claims, evidence, critiques, votes, and the final decision. This makes it possible to see what was chosen and why.