bug: task-orchestrator and task-executor agents break user commands #165

New issue

Open

opened 2025-10-14 15:44:25 -06:00 by navan · 0 comments

navan commented

2025-10-14 15:44:25 -06:00

Owner

Originally created by @svallory on 8/17/2025

Description

Telling Claude to always use certain agents to execute a tasks by adding the task-orchestrator.md and task-executor.md broke several of my commands. The issues were:

these agents are picked over other more specialized agents for the task at hand
In Claude code, subagents cannot start other subagents. so parallelism is completely gone

Steps to Reproduce

Have a typescript-expert agent
Run a typescript task
task-executor will be used

Expected Behavior

Do not add new agents without telling the user about them
Better scope the agent execution.
Instead of

Use this agent when you need to implement, complete, or work on a specific task that has been identified by the task-orchestrator or when explicitly asked to execute a particular task...
Use something like:
When running a tm:* claude command, unless a more appropriate agent exists, use...
Have the agent ASK the user if they should run the task-orchestrator and warn them this is slow and there will be no parallel agent execution

Another option is to convert these agents to commands. In fact, there are already commands for these, I don't see any need for these agents.

Actual Behavior

Already described

Environment

Task Master version: 0.24.0
Operating system: Mac OS
IDE (if applicable): terminal / claude code

*Originally created by @svallory on 8/17/2025* ### Description Telling Claude to always use certain agents to execute a tasks by adding the `task-orchestrator.md` and `task-executor.md` broke several of my commands. The issues were: - these agents are picked over other more specialized agents for the task at hand - In Claude code, subagents cannot start other subagents. so parallelism is completely gone ### Steps to Reproduce 1. Have a typescript-expert agent 2. Run a typescript task 3. task-executor will be used ### Expected Behavior 1. Do not add new agents without telling the user about them 2. Better scope the agent execution. Instead of > Use this agent when you need to implement, complete, or work on a specific task that has been identified by the task-orchestrator or when explicitly asked to execute a particular task... Use something like: > When running a `tm:*` claude command, unless a more appropriate agent exists, use... 3. Have the agent ASK the user if they should run the task-orchestrator and warn them this is slow and there will be no parallel agent execution Another option is to convert these agents to commands. In fact, there are already commands for these, I don't see any need for these agents. ### Actual Behavior Already described ### Environment - Task Master version: 0.24.0 - Operating system: Mac OS - IDE (if applicable): terminal / claude code