analyst
Pre-planning requirements agent that converts product scope into concrete, testable acceptance criteria.
The analyst agent works the gap between a raw user request and the planning stage. It reads what was actually asked, drags every unanswered question into the open, and writes acceptance criteria you can objectively test. No guessing. That kills the "but I thought you meant..." conversation before implementation starts instead of after.
Role
- Identify missing questions, undefined guardrails, and unvalidated assumptions in a request
- Surface scope risks and edge cases (unusual inputs, race conditions, boundary values)
- Produce acceptance criteria that are objectively pass/fail rather than subjective
- Define explicit scope boundaries — what is included and what is excluded from the work
When invoked
| Situation | How it's triggered |
|---|---|
Before planner at the start of $autopilot and $plan | Automatic |
During $deep-interview when the goal itself is ambiguous | Automatic |
| When a task description is vague, contradictory, or missing success criteria | Direct request |
| Requirements audit before executing an existing plan | Direct request |
Inputs
- The user's task description or feature request
- Optional: a prior exploration report from the
exploreagent - Optional: referenced specification documents or existing acceptance criteria
Outputs
A structured analysis document written to .omx/specs/<date>-<topic>.md containing:
- Missing questions ranked by impact
- Undefined guardrails with suggested concrete bounds
- Scope risks with mitigation strategies
- A validated assumption list with verification methods
- Testable acceptance criteria
Limits
- Does not design the implementation — that is the
architectagent's responsibility - Does not write or modify any production or test code
- Does not prioritize features by market or user value — focuses only on implementability and completeness