How to use Claude skills and subagents with Task-master?

[ZeFifi]

Hello everyone,

I'm running into an issue when using Task-master with Claude Code: my agents and skills never seem to activate. I updated the Claude.md file in the Task-master folder accordingly, but it still doesn’t work.

Here’s my usual workflow, in case I’m missing something:

• run task-master add-task --prompt="my prompt" in the terminal
• break the task down into subtasks
• start working through them with Claude

My Claude.md file currently contains:

### Iterative Implementation

1. `task-master show <subtask-id>` - Understand requirements
2. Explore codebase and plan implementation
3. `task-master update-subtask --id=<id> --prompt="detailed plan"` - Log plan
4. `task-master set-status --id=<id> --status=in-progress` - Start work
5. Implement code following logged plan with relevant Claude skills (`~/.claude/skills/`) and subagents (`~/.claude/agents/`) to ensure the code is correct and follows best practices (react-best-practices, typescript-best-practices, tailwind-guidelines, accessibility-auditor and so on).
6. `task-master update-subtask --id=<id> --prompt="what worked/didn't work"` - Log progress
7. `task-master set-status --id=<id> --status=done` - Complete task

Could you help me figure out what I might be doing wrong?

Thanks,
Philippe

[Crunchyman-ralph]

hey philippe, thanks for your message, can you maybe clarify some things for me.

what are you expecting to happen, what is actually happening?

[ZeFifi]

Hey Ralph, thanks for getting back to me so quickly. I was expecting my Claude skills or sub‑agents to kick in when needed. For instance, I was doing a TypeScript refactor this morning, and I actually have a Claude Skill for that. I thought the typescript-best-practice skill would trigger, but it never did. Claude takes care of the work by itself without skills or subagents.

What I want is for Claude to trigger a global skill or sub‑agent (located in ~/.claude/) automatically when I’m working with Task‑master.

[Crunchyman-ralph]

hmmm, in that case maybe it makes sense to be specific inside of the task details?

maybe in your PRD you can specify your skills as context, in that case while doing the task deconstruction it will trigger a skill or an agent.

honestly I haven't yet dug into skills, so i don't really understand the logic behind how it's triggered

[ZeFifi]

Yes, maybe in the task details. I'll try it and keep you posted! Thanks!

[romanr]

maybe in your PRD you can specify your skills as context, in that case while doing the task deconstruction it will trigger a skill or an agent.

I got agents configured with instructions about frameworks and best practices for the agent's particular area, such as testing, frontend or backend. Also includes MCP tool usage instructions that are common for most but not all agents, typically Tester use different MCP services and developers their own set of MCP.
Each sub-agent has its own configuration file. plus there's referenced common MCP instructions files.

After PRD is parsed, tasks created, then I should update every task's description by copying and pasting relevant agent configuration? And then ask to re-evaluate tasks using these new details?

Maybe I am not understanding something, but that goes against the whole idea of having an agent - that skills are separated from the task.

I imagine the specialised configured agent could be passed as an executor for a task.

[Crunchyman-ralph]

@romanr were thinking of an orchestration layer at the moment and what you're saying makes a lot of sense.

specialised agent for specific tasks !

[ZeFifi]

After several attempts, the best way to use an agent with Taskmaster is to call it explicitly in the prompt to start a task. For example: “start task X with agent Y.” I’ve never managed to trigger it automatically, but this method works!

[Crunchyman-ralph]

Nice, thats what I thought!