New: watch your coding agents work, step by step, on a live board. Try the demo, no account

Run several coding agents in parallel on one task list

By Jake Marsh, who builds Taskhold. Updated 2026-10-31. 4 min read.

Running agents in parallel, in separate worktrees or separate tools, is common now. The hard part is not starting them. It's keeping them off each other's work and knowing what each one finished. A shared list with a live board solves both: each agent claims a task before it starts, and the board shows every agent's progress side by side. The CLAUDE.md task workflow is what makes agents claim tasks, see what each agent is doing covers watching them, and the live board demo shows it running. For just two tools, see one todo list for Claude Code and Cursor; for agents spread over several repos, see one task list across repos. How common this has become: AI coding statistics.

The pattern

  • One list for all agents, outside any single repo or session.
  • Each agent claims a task (status in progress) before it starts.
  • Agents only take todo tasks, so a claimed task is skipped.
  • Agents block with a reason instead of guessing, and complete with a summary.
  • You review blocked tasks and summaries in one place.

Set it up with Taskhold

Make one setup URL per agent or tool in Settings, so each has its own key you can revoke. Paste each into its agent. Add the workflow snippet to CLAUDE.md or AGENTS.md in each repo:

## Tasks (Taskhold)

The person and I share a task list through the taskhold MCP tools. They watch a live board while I work.
Pass project: "<repo folder name>" on every call (created on first use, no ids needed).
- Session start: taskhold_get_current and taskhold_updates, both with project. Keep the cursor.
- New request: taskhold_begin with title, project, steps and agent: "<client> in <repo>". It files the task, plans the steps and starts the first one. Queued work: taskhold_next with project and agent takes the next To Do task and starts it; after each one, call it again until it says To Do is empty.
- Project notes in a reply (stack, commands, what done means, what to avoid) are the person's rules for every task in that project.
- Each step done: taskhold_complete with start_next: true. Steps drive the progress bar.
- Long step: taskhold_progress on the step (N of M, or a percent) with a short message. Every few minutes at most.
- Need a decision: taskhold_block with one specific question, then taskhold_updates with task_id, since and wait: 50 until the reply arrives.
- Done: taskhold_complete on the task with what changed and how I checked it.
- A TODO or bug I'm not fixing now: taskhold_create with project (or taskhold_create_from_todo with file and line).

If agents work in different repos, have each pass its repo folder name as project ("Pass project: <repo folder name> on every taskhold call"). Projects are created on first use.

Watch the agents side by side

Parallel agents are where a progress view pays off: three terminals tell you little at a glance. On the live board each running task gets a card with the agent's label, a bar split by step, the time it has run and its latest message, and blocked tasks from any agent collect at the top. A card whose dot reads No word has not reported for fifteen minutes, which is usually the one to look at first.

Limits to know

  • Claiming is a status change, not a lock. Two agents that start the same task in the same second can both get it.
  • Which agent did something comes from the label it passes to taskhold_start, or the client name its requests carry. Keys are per client, not per session, so give parallel sessions of one client different labels.
  • There are no shared team workspaces yet: the list belongs to one person and their agents.
Share this page:XLinkedInBlueskyHacker News

Answer your coding agents from anywhere.

$12/month or $120/year after the trial. Card up front. Cancel before day 7 and you pay nothing. Every agent you run is included.

Or connect your agent first, no sign-up: add https://taskhold.com/api/mcp as an MCP server. The first time it uses Taskhold, it gives you the trial link.

Questions

How many agents can connect?
There is no per-agent limit or charge. The plan is per person and covers every agent that person runs.
Do agents see each other's notes?
Yes. taskhold_get returns a task with all its notes, so an agent picking up a blocked or half-done task sees what the last one wrote.
What does it cost?
$12 a month or $120 a year per person, with a 7-day trial that starts with a card.

Related

Updated 2026-10-31. Something wrong or out of date? Email hello@greatwork.company.