Summary
AgentFlow currently presents an agent-oriented task board that is closer to a lightweight Kanban/Trello model. As agent tasks become longer-running, multi-step, and multi-agent, the board would benefit from a small set of Jira-inspired semantics without inheriting Jira's full complexity.
Proposal
Introduce a Jira-lite, agent-native issue model for task boards:
- Issue types: task, bug, story, sub-task, incident, research
- Agent-aware statuses: backlog, ready, assigned, running, waiting for approval, blocked, review, done, failed
- Structured task fields: priority, reporter, assigned agent, environment, success criteria, risk level, approvals, artifacts, runtime/cost metadata
- Parent/child and dependency relationships for spawned work and blocking work
- Event timeline integration so agent session events become issue history
Why this matters
A plain Kanban board is useful for visibility, but agent workflows need stronger task semantics:
- Agents need explicit success criteria to know when to stop.
- Humans need an approval lane for risky external actions.
- Multi-agent execution needs dependencies and child tasks to avoid duplicated or conflicting work.
- Long-running sessions need durable timeline/history instead of scattered logs.
Suggested first implementation
Start with a minimal schema and UI/API support for:
- Issue type
- Priority
- Success criteria
- Risk level
- Dependency links
- Agent-native statuses including
waiting_for_approval, blocked, review, and failed
This should preserve the existing lightweight board while adding enough structure for more reliable agent coordination.
Summary
AgentFlow currently presents an agent-oriented task board that is closer to a lightweight Kanban/Trello model. As agent tasks become longer-running, multi-step, and multi-agent, the board would benefit from a small set of Jira-inspired semantics without inheriting Jira's full complexity.
Proposal
Introduce a Jira-lite, agent-native issue model for task boards:
Why this matters
A plain Kanban board is useful for visibility, but agent workflows need stronger task semantics:
Suggested first implementation
Start with a minimal schema and UI/API support for:
waiting_for_approval,blocked,review, andfailedThis should preserve the existing lightweight board while adding enough structure for more reliable agent coordination.