Sometimes teams stop claiming tasks

[CKGrafico]

What happened?

I dont find the pattern but sometimes after having a lists of tasks in ensemble the agents start to spawnand claim and implement taks and then stop claiming next ones until i ask to do it, is this common?

Steps to reproduce

1. create team
2. assign agents
3. assign tasks
4. ask to start the loop of claming and implementing

Relevant logs or errors

I'm using windows (no wsl)

Plugin version

0.5.0

OpenCode version

1.2.27

Operating system

Other

[hueyexe]

Thanks for the report. 0.5.0 is quite old, and several task/reporting/stall-handling fixes landed after that version.

Please update to the current pinned plugin version, restart OpenCode, and try the same flow again:

@hueyexe/opencode-ensemble@0.15.0

If it still reproduces, please share:

• the exact plugin version after update
• what team_status shows when agents stop
• whether remaining tasks are pending, blocked, or in_progress
• whether the agents sent team_message and called team_tasks_complete

[CKGrafico]

Sorry I wrote the version wrong.. I'm using the newwest one
Let me show you a sample of the log

team_spawn [name=front2, agent=frontend-engineer, prompt=You are front2, frontend-engineer on this team.

Initial batch of tasks:
- [task_mq2b2ytd_0001_2e8pbx] 1.1 Redesign `src/app/[locale]/page.tsx` from the current alpha-centered stack into a beta-oriented public homepage with a clear hero, value messaging, and visible join-beta CTA

Key context:
- Change: beta-landing-page, spec-driven, public locale root homepage only
- Keep root route public for unauthenticated visitors, no app-shell chrome
- Reuse existing waitlist flow; do not change API contract or DB model
- This task must unblock later work on beta-join form copy and multilingual messages across 9 locales
- Deliver a responsive marketing-style beta landing page with strong value framing and clear CTA placement

Available OpenCode tools:
- team_claim
- team_tasks_complete
- team_tasks_list
- team_tasks_add
- team_message
- team_broadcast

How to proceed:
- FIRST action: call team_claim with exact task_id `task_mq2b2ytd_0001_2e8pbx`
- After finishing that task, call team_tasks_complete with exact task_id `task_mq2b2ytd_0001_2e8pbx`
- Then message lead with files changed, checks run, and whether you are ready for next tasks
- Lead may assign more tasks. Do not shut down until lead confirms no more tasks

Before starting, read and follow these skills:
- `.agents/skills/ob-global/SKILL.md`
- `.agents/skills/react-nextjs-patterns/SKILL.md`
- `.agents/skills/tailwind-shadcn/SKILL.md`
- `.agents/skills/guardrails-habflow/SKILL.md`

Use model rules already in session. Use `rtk` for shell commands covered by guardrails., model=openai/gpt-5.4, claim_task=task_mq2b2ytd_0001_2e8pbx, worktree=true, plan_approval=false]
team_message [to=front2, text=Start now. Force claim first. Your task: [task_mq2b2ytd_0001_2e8pbx] 1.1 Redesign `src/app/[locale]/page.tsx` from the current alpha-centered stack into a beta-oriented public homepage with a clear hero, value messaging, and visible join-beta CTA. First action: call team_claim task_id:task_mq2b2ytd_0001_2e8pbx before any other work., approve=false, reject=]
Done.
Current setup:
Stalled front agent force-shut down
Replacement spawned: front2
Same task assigned: [task_mq2b2ytd_0001_2e8pbx]
Model: openai/gpt-5.4
Start message explicitly requires claim-first behavior
Note:
There is no separate “force claim” tool on the lead side
Strongest control available is:
- spawn with claim_task:"task_..."
- send a start message that makes team_claim the first action
- recycle stalled agents immediately
Waiting for front2 now.
▣  Build · GPT-5.4 · 56.6s
done?
+ Thought: Checking answer status · 1.9s
Checking whether front2 actually started and whether any task completed.
team_status 
team_tasks_list 
team_results [from=front2]
+ Thought: Considering options for process improvement · 2.4s
No.
Current status:
front2 is also stalled
No messages
No task claimed/completed
Board unchanged: 1 pending, 6 blocked
Pending:
[task_mq2b2ytd_0001_2e8pbx] 1.1 redesign src/app/[locale]/page.tsx
Assessment:
The ensemble worker path is failing to start work in this session
Nothing has been implemented yet
Next move:
force shutdown front2
respawn a new worker, likely basic-engineer, on openai/gpt-5.4 with the same task and explicit claim-first instruction
If you want, I’ll do that now.
▣  Build · GPT-5.4 · 15.2s
force
Forcing another recycle now. I’m shutting down front2 and replacing it with a fresh basic-engineer worker on the same task.
team_shutdown [member=front2, force=true]
team_spawn [name=builder1, agent=basic-engineer, prompt=You are builder1, basic-engineer on this team.

Initial batch of tasks:
- [task_mq2b2ytd_0001_2e8pbx] 1.1 Redesign `src/app/[locale]/page.tsx` from the current alpha-centered stack into a beta-oriented public homepage with a clear hero, value messaging, and visible join-beta CTA

Key context:
- Change: beta-landing-page, spec-driven, public locale root homepage only
- Keep root route public for unauthenticated visitors, no app-shell chrome
- Reuse existing waitlist flow; do not change API contract or DB model
- This task must unblock later work on beta-join form copy and multilingual messages across 9 locales
- Deliver a responsive marketing-style beta landing page with strong value framing and clear CTA placement

Available OpenCode tools:
- team_claim
- team_tasks_complete
- team_tasks_list
- team_tasks_add
- team_message
- team_broadcast

How to proceed:
- FIRST action: call team_claim with exact task_id `task_mq2b2ytd_0001_2e8pbx`
- After finishing that task, call team_tasks_complete with exact task_id `task_mq2b2ytd_0001_2e8pbx`
- Then message lead with files changed, checks run, and whether you are ready for next tasks
- Lead may assign more tasks. Do not shut down until lead confirms no more tasks

Before starting, read and follow these skills:
- `.agents/skills/ob-global/SKILL.md`
- `.agents/skills/react-nextjs-patterns/SKILL.md`
- `.agents/skills/tailwind-shadcn/SKILL.md`
- `.agents/skills/guardrails-habflow/SKILL.md`

Use model rules already in session. Use `rtk` for shell commands covered by guardrails., model=openai/gpt-5.4, claim_task=task_mq2b2ytd_0001_2e8pbx, worktree=true, plan_approval=false]
team_message [to=builder1, text=Start now. Force claim first. Your task: [task_mq2b2ytd_0001_2e8pbx] 1.1 Redesign `src/app/[locale]/page.tsx` from the current alpha-centered stack into a beta-oriented public homepage with a clear hero, value messaging, and visible join-beta CTA. First action: call team_claim task_id:task_mq2b2ytd_0001_2e8pbx before any other work., approve=false, reject=]
Forced.
Current worker:
builder1
Agent: basic-engineer
Model: openai/gpt-5.4
Task: [task_mq2b2ytd_0001_2e8pbx]
What changed:
front2 force-shut down
fresh worker spawned
explicit claim-first start message sent
Waiting for builder1 to claim and report.

and then nothing again, I've a lot of examples like that