Task Lifecycle

OpenViber supports two task origins:

• Self-initiated (agent discovers and proposes/picks work),
• Manager-assigned (explicit human request).

Both follow the same lifecycle with human checkpoints.

1. Lifecycle states

2. State intent

• Intake: classify origin (self-initiated or manager-assigned) and execution mode (Always Ask, Agent Decides, or Always Execute).
• Planning: update workspace/task.md and next actions.
• Executing: run tools, terminals, browser tasks, code edits.
• Verifying: validate outcomes from human-observable perspective.
• Reporting: emit report with evidence refs.
• AwaitingFeedback: ask manager for decisions (prefer MCQ).
• Paused: wait due to policy, budget, or manager instruction.
• Blocked: requires account access/approval/decision.

3. Work products by phase

• Planning -> plan diff, priorities, budget estimate.
• Executing -> terminal traces, intermediate artifacts.
• Verifying -> screenshots/logs/checklist with pass/fail.
• Reporting -> concise status + links to proof.

4. Periodic reporting

For long tasks or self-initiated work, vibers should report on cadence (chat/email), including:

• what changed since last report,
• current risks/blockers,
• next planned actions,
• budget burn vs limit.

5. Human feedback requests

When blocked by preference ambiguity, ask with options:

• recommended option first,
• clear tradeoffs per option,
• one question at a time unless tightly coupled.

6. Completion criteria

A task is complete only when:

1. acceptance checks pass from human perspective,
2. report includes verification clues,
3. relevant plan/progress/artifacts are persisted in workspace.