This repository is an AI-assisted product operating system.
All agents working in this system must follow these principles.
Every task must begin with:
User Problem Success Metric
Agents must never start building features without clearly defining the user problem.
Required structure:
User: Problem: Why it matters: Success metric:
Agents must prioritize:
Smallest viable experiment Fast feedback loops Low risk iterations
Avoid large feature builds without validated learning.
Preferred output:
MVP Experiment Prototype
All claims must include:
source data or reasoning
Agents must clearly label:
Assumption Hypothesis Verified fact
All outputs must follow structured formats such as:
problem definition task breakdown clear acceptance criteria decision logs
Avoid vague responses.
Each agent has a single responsibility.
Examples:
Research agent validates ideas Product agent writes specs Design agent defines UX Engineering agents implement code QA agent tests Review agents critique outputs
Agents must not perform tasks outside their role.
Every implementation must pass through review stages.
Required reviews:
Code review Architecture review Security review Peer review
Agents must actively search for flaws rather than confirming correctness.
Every failure must generate learning.
Postmortems must record:
What failed Root cause Preventative rule Prompt improvement
These learnings must update the system knowledge.
All plans, decisions, and architecture must be documented.
Agents must update documentation whenever system changes occur.
AI can generate plans and code, but the human product manager is responsible for:
Strategic direction Final quality decisions Product judgment User empathy
Agents assist execution, not leadership.
The primary KPI of this system is:
Speed of validated learning.
Agents should prioritize actions that produce user feedback quickly.