Task line: tasks pass from agent to agent in order

The default line has four stages. You can switch each stage’s agent between Claude and Codex, and a line can have up to 8 stages.

• Analysis: writes its findings to 01-analiz.md.
• Plan: writes the plan to 02-plan.md.
• Build: writes the code.
• Review: marks each acceptance criterion with evidence and approves or rejects the work.
• Each task’s files sit in their own folder under .agentvera/hat/.

You write the task as plain text; if it holds several jobs they become separate tasks, each with acceptance criteria. A stage ends its work with the [[TESLİM]] marker. If the reviewer rejects, the task goes back to Build with the findings; after the number of rounds you set (3 by default) it is flagged as needing attention. An agent that stops without a marker is nudged twice, then you are told.

• The reviewer marks the criteria, not the agent that wrote the code; nothing passes without evidence.
• Task states: queued, working, done, needs attention, stopped.
• You can reorder queued tasks, accept a task as it is, stop it or remove it.
• When a task is done, a desktop notification says how many criteria passed.

A task’s verification view shows its acceptance criteria as passed, failed and pending, each with its evidence note, plus the task’s history across stages. If a stage’s agent waits for approval or closes, the task is flagged as needing attention; you open the agent, look, then let the task continue.