Guide To Evaluator Workflow

Sinaris has two entry paths and one shared delivery loop.

Guide

Use Guide when a project is new or the direction is unclear.

Guide establishes the project basis: goal, users, business value, scope, decisions, open questions, and milestone roadmap. The output is durable context, not a disposable chat summary.

Proposal

Use Proposal when an existing project receives a concrete new requirement, change, or idea.

Proposal shapes one demand at a time. It captures the problem, observed context, assumptions, options, recommendation, risk, and context action before the work enters design.

Design

Design turns confirmed direction into executable phase work.

It produces phase design markdown and ready tasks. Apply should not invent missing product or architecture decisions. If the design is insufficient, the workflow returns to Guide, Proposal, or Design.

Apply

Apply is the implementation loop.

Agents claim one task at a time, read the phase context and work packet, implement within scope, verify the result, and complete or fail with evidence.

If a task needs a user decision, the agent runs the decision flow instead of guessing. If another task is claimable after completion, the Apply SKILL decides whether to continue.

Activity

Activity is the trace of runtime handoffs.

It combines session events, task logs, structured AgentEvent items, and optional transcript references so humans can understand what happened and why an agent resumed.

Evaluator

Evaluator tracks delivery confidence beyond task completion.

It asks whether important user journeys still work and whether the available evidence supports shipping.

Hub

The Hub is the human runtime view.

It shows plans, task readiness, agent sessions, structured activity, decisions, workspaces, and evaluator confidence without requiring users to reconstruct state from chat history.