Hand tasks to your coding agent and get answers back
By Jake Marsh, who builds Taskhold. Updated 2026-10-26. 4 min read.
Most of us hand work to an agent by typing it into the chat. That works for the task in front of you. It doesn't work for the five things you thought of on a walk, or for knowing which of yesterday's requests got done.
A better pattern is a queue: you write tasks down when you think of them, the agent takes them in order, and each task comes back with either a summary or a question. With Taskhold, the live board shows the task the agent is on, step by step, while it works. For a big piece of work, break the spec into tasks first; when the run is over, review what the agent did on the tasks themselves; and if it stops to ask, answer it from the board.
Write tasks an agent can finish
- One outcome per task: "Add rate limiting to /api/login (5 per minute per IP)" rather than "security stuff".
- Say when it's done. List the checks you'll run yourself: a test that passes, a curl that returns 429, a page that shows the new column. The agent runs them before it calls the task done.
- Say what's out of scope. "Don't touch the signup flow" saves a diff you'd throw away.
- Point at an example: the file or PR to copy, and one you don't want copied.
- Set priority (the agent takes urgent, then high, then medium, then low), and group by project if you work in several repos.
- End with "If unsure, block with one question." The question shows up on your live board instead of a guess in the code.
A template you can paste into the task description:
Why: <one line: what this unblocks or who asked>
Do: <the outcome, one sentence>
Done when:
- <check 1 you can run: a test, a curl, a page that shows X>
- <check 2>
Not in scope: <what to leave alone>
Look at: <files, links, an example to copy>
Don't copy: <an example of what you don't want, if any>
Constraints: <no new dependencies, keep the public API, etc.>
If unsure: block with one question; don't guess.Let the agent pick up the next one
With Taskhold connected over MCP and this snippet in CLAUDE.md or AGENTS.md, "work on the next task" is the whole prompt:
## 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).Blocked is an answer too
When the agent needs a decision, a credential or a review, it calls taskhold_block with a specific question and moves on. The task goes to the top of the live board under Needs you. Type the answer and press Send and unblock; the agent reads it with taskhold_updates the next time it checks, instead of you finding the question buried in a terminal.
Add tasks from anywhere
The web app works on a phone. Add a task while you're away from the desk, and the agent sees it at the start of its next session.
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
- Which agents work?
- Claude Code, Cursor, Codex, Windsurf (Devin Desktop) and any MCP client that supports remote (Streamable HTTP) servers with a header.
- Does the agent run on its own when I add a task?
- No. Taskhold is the list, not a runner. The agent sees new tasks the next time it checks, for example at the start of a session.
- What does it cost?
- $12 a month or $120 a year per person, with a 7-day trial that starts with a card.
Related
- Answer a blocked coding agent from your phone
- Break a spec into tasks for your coding agent
- What your coding agent did while you were away
- A CLAUDE.md and AGENTS.md task workflow
- Claude Code task list over MCP: see progress live
- Run Claude Code or Codex overnight from a queue
- Is it safe to connect Claude Code to a task list?
- File a Taskhold task when GitHub Actions fails
Related guides from other Great Work products
Updated 2026-10-26. Something wrong or out of date? Email hello@greatwork.company.