Feature Discovery

Start with a delegation idea
Give AI Analyst a delegation goal and REQ-1 as trusted context. Plan mode asks questions before proposing specifications.
Discovery Before Coding
Start with the existing product context and resolve decisions that affect the feature.
Start from context
Use a module, requirement, data entity, UI, software unit, or issue.
Investigate impact
Ask AI Product Expert when the team needs references, source files, risks, or recommended next actions before changing specs.
Clarify decisions
AI Analyst asks only necessary questions, proposes answer options, and stores an execution brief for the spec pass.
Review spec changes
Inspect diffs, edit descriptions, adjust proposed items, and keep AI output separate from product truth until Apply.
Apply under a CR
Link the AI Analyst changes to a Change Request and apply the reviewed result in the main or feature branch.
Hand off Change Request
The coding agent reads the Change Request, affected specs, AI Analyst summaries, app manifest, architecture guidelines, and file traces through MCP.
AI Analyst narrows uncertainty before it writes specs
AI Analyst reads the available context, asks focused questions, and offers options when the context supports them. It records facts, assumptions, artifact keys, and a brief for the specification pass.
When enough detail is known, it stores source facts, assumptions, relevant artifact keys, and a concrete execution brief for the specification pass.
Example clarification questions
- Which actors can approve enterprise onboarding?
- Should rejected onboarding return to draft or become terminal?
- Should the API expose approval state in the existing endpoint or a new operation?
- Which UI transition should notify the requester?
AI Analyst output is not product truth until humans apply it
Review proposed requirements, data, UI, wireframes, APIs, and justified unit links as a diff. Edit or disable result items, then apply the accepted changes explicitly.
The result is diffable, editable, and scoped to the actual changes. Users review the proposed deltas instead of re-reading entire specifications: they can refine proposed descriptions and then apply the effective result explicitly.
Choose the Right Starting Point
Feature discovery can start from an idea, an existing spec, a Change Request, code, or a detected mismatch. The end state is the same: implementation-ready context.
Idea first
Use AI Analyst clarification when the desired outcome is known but behavior, edge cases, and contracts are still open.
AI Product Expert first
Use AI Product Expert when the team needs to understand current specs, source files, and likely impact before deciding what to change.
Change Request first
Create the work package first when the feature is already approved and needs branch scope, review history, and one handoff key.
Known spec first
Start from REQ-*, UI-*, API-*, D-*, module, or unit when the affected artifact is already clear.
Code first
Trace a source file to units and specs, then decide whether the intended change belongs in specifications, implementation, or both.
Findings- and Issues-driven
Start from system-detected findings when verifier evidence shows a mismatch, or from human-created issues when people raise ideas, concerns, bugs, or other reasons for change.
Continue from accepted specifications
The reviewed Change Request is the starting point for feature implementation. If later checks reveal a specification gap, return to discovery with the evidence.
If the implementation exposes an incomplete or contradictory spec, the team can continue discovery under the same Change Request or create the next one with evidence.
Before and After Reqode
Resolve behavior and review specifications before the coding agent edits files.
Without Reqode
- The feature brief lives in chat, a ticket, or a prompt.
- AI coding starts before product decisions are clear.
- Specifications are updated after implementation or not updated at all.
- Reviewers reconstruct why code changed.
With Reqode
- Discovery happens before implementation.
- AI Analyst asks focused questions and proposes options.
- Specification changes are reviewed before Apply.
- The coding agent receives
CR-123and reads focused and traceable context through MCP.
Build with discovery, not guesswork
Clarify one feature, review the changes, and leave discovery with an accepted scope.



