Skip to content
Agentmart
Submit a tool

tasks

Defer, track, and resume multi-step work with task-create/list/update/complete, plus plan and checkpoint primitives.

Skill by josephfung

IndexedMIT★ 9
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/josephfung-tasks

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Defer, track, and resume multi-step work with task-create/list/update/complete, plus plan and checkpoint primitives. Pin this skill to make an agent heartbeat-eligible.

Use cases: productivity, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Node.js package

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/tasks

What's inside

Skill name
tasks
Description
Defer, track, and resume multi-step work with task-create/list/update/complete, plus plan and checkpoint primitives. Pin this skill to make an agent heartbeat-eligible.
License
MIT
Instructions
39 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Files (19)

  • SKILL.md
  • handler.test.ts · tools/checkpoint/
  • handler.ts · tools/checkpoint/
  • tool.json · tools/checkpoint/
  • handler.test.ts · tools/plan/
  • handler.ts · tools/plan/
  • tool.json · tools/plan/
  • handler.test.ts · tools/task-complete/
  • handler.ts · tools/task-complete/
  • tool.json · tools/task-complete/
  • handler.test.ts · tools/task-create/
  • handler.ts · tools/task-create/
  • tool.json · tools/task-create/
  • handler.test.ts · tools/task-list/
  • handler.ts · tools/task-list/
  • tool.json · tools/task-list/
  • handler.test.ts · tools/task-update/
  • handler.ts · tools/task-update/
  • tool.json · tools/task-update/

SKILL.md (first lines)

## Task Management

You can defer, track, and resume work using your task skills.

**Decide, don't drop.** When work arrives that you cannot finish now, create a
task (`task-create`, optionally with `wake_at`) rather than cramming it into one
burst or abandoning it. Briefly tell the principal what you queued and why.

**Own the how.** When a clear goal is too big for one burst, decompose it and start —
with a one-line heads-up. Do not ask the principal how to execute (splitting, scheduling,
budgets); ask only about the goal itself, or a consequential or hard-to-reverse action.

**Decompose projects.** If work has more than one step, or any step cannot be done
right now, create a parent task whose `intent_anchor` states the durable goal, plus
the first wave of subtasks (`parent_task_id`, and `blocked_by_task_id` for ordering).
Plan the first wave only; add subtasks as you learn more.

**Make every subtask self-contained.** A woken subtask starts fresh with no memory of this
conversation — it has only its id, title, intent, and progress note. So its
`intent_anchor` and progress must fully re-establish the work: the goal, the resource
(e.g. a URL), and precise progress ("answered Q1-12 of 60"). A bare title like "Answer
Q13-36" strands the wake. For **web-browser** work, also record the exact `session_id` and
whether the session was `incognito`, and resume it with the same `incognito` value.

**Advance until blocked.** When you act on a task, do every step you can right now.
Stop only at a real blocker — waiting on a person, on the principal's approval, on a future
date, or on a prior task — or when your turn budget runs low. Then park each loose end:
set its status (`waiting`/`blocked`), add a progress note, and set a wake (a reply you
are expecting, or a `wake_at` timer). When a task's goal is achieved, call `task-complete`
with a note.

**Never promise without a task.** Before you send anything that commits to a future
action ("I'll follow up with X", "we'll send that over"), make sure a task backs that
promise. Prefer to resolve the dependency first and send a complete message. Only send
an interim "I'll follow up" when the recipient needs an acknowledgment now — and when
you do, create the follow-up task (yours if you can chase it; the principal's if only they
can, and tell them).

From josephfung/curia, MIT.

More from josephfung/curia

Signals

Stars
9
Forks
6
Open issues
67
Last commit (this folder)
Oct 13, 2026 (16 days ago)
Last release
Oct 16, 2026 (12 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from josephfung

Cite this page

tasks skill. Agent Mart. https://agentmart.dev/skills/josephfung-tasks (last updated 2026-10-24)