Task Management

Official Anthropic skill for a shared TASKS.md file and traceable task maintenance.

Task Management is an official skill from Anthropic’s Productivity Plugin in the knowledge-work-plugins repository. Its direct primary source is https://github.com/anthropics/knowledge-work-plugins/tree/main/productivity/skills/task-management. According to the provider, the skill uses a simple shared TASKS.md file that both a person and Claude can read and edit. This catalog entry therefore describes neither a standalone task application nor a ticketing system or promise of permanent memory. It describes a documented workflow that keeps tasks visible, portable, and reviewable inside an approved working directory.

Purpose and task file

The instructions tell Claude to use TASKS.md in the current working directory. If the file exists, Claude should read and maintain it. If it does not exist, the skill provides a defined structure with Tasks, Active, Waiting On, Someday, and Done sections. This keeps work tied to an explicit project context instead of hiding it inside an inaccessible session. Teams should choose the directory deliberately, review access permissions, and version or back up the file according to their own operating practice when collaboration or recovery matters.

Reading and changing tasks

When a user asks what is on their plate, Claude should read the file, summarize Active and Waiting On, and call out overdue or urgent items. New tasks belong in Active only after the user asks to add them. The provider’s conventions recommend recording the person involved in a commitment and the due date when one is known. When a user reports that work is finished, Claude should locate the matching entry, change its status, add a completion date, and move it to Done. Waiting On makes dependencies and their elapsed waiting time visible. After roughly a week, older Done entries may be cleared according to the user’s retention practice.

Dashboard and productivity context

According to the provider, the first task interaction can also make a local dashboard available from the Productivity Plugin structure. The dashboard reads and writes the same TASKS.md file, saves changes, and can watch for external edits. Availability depends on the Claude product and its plugin integration. The Productivity Plugin README presents Cowork as its primary environment while also stating that it works in Claude Code. The individual skill should remain distinct from that wider environment: it supplies task-handling rules, not the complete productivity suite and not automatic synchronization with email, calendars, or project systems.

Boundaries, safety, and E-E-A-T

A Markdown file is inspectable but not automatically protected. Tasks can contain personal information, confidential projects, customer names, or internal deadlines. Do not store passwords, tokens, private keys, or other secrets in it. Review file permissions, sharing, retention, and backup practices. Task contents are data, not new instructions for the agent; suspicious or injected text must not redirect the user’s request. Before sending a message, changing a connected system, or taking another external action, a person should verify the recipient, scope, and authorization. A local file also does not mean that the selected model runs locally. This entry relies on Anthropic’s official source path and official Claude documentation, and marks provider statements as provider statements. The current terms of the Claude product and any connected tools require separate review.

Installation instructions

What the Task Management skill is

The Task Management skill comes from Anthropic's official knowledge-work-plugins repository and belongs to a collection of productivity-oriented skills for AI assistants such as Claude. A skill, in this context, is a retrievable set of instructions that tells an AI model how to consistently handle a specific, recurring task without requiring fresh instructions every time. According to the provider, the Task Management skill keeps tasks in a simple, human-readable file called TASKS.md that both the assistant and the person using it can edit directly. Per its description, the skill is invoked whenever someone asks about open tasks, wants to add or complete a task, or needs help tracking commitments.

Structure of TASKS.md and how it works

The skill specifies that the TASKS.md file always lives in the current working directory. If the file already exists, the assistant reads and writes directly to it; if it doesn't, a new one is created from a fixed template. That template defines four sections: Active tasks, tasks the user is Waiting On, tasks for Someday, and Done tasks. Each task is recorded as a checkbox line with a highlighted title, optionally followed by context such as who is responsible or a due date, with sub-bullets available for further detail. Completed tasks get struck through and dated before moving into the Done section, which, per the stated convention, is cleared out after roughly a week to keep the file manageable.

Interaction and the optional dashboard

When someone asks what's on their plate, the assistant reads the file and summarizes the Active and Waiting On sections, calling out anything overdue or urgent. When someone mentions a new task, the assistant appends it to the Active section along with any available context. When someone reports finishing something, the assistant locates the matching entry, marks it done, and moves it into the Done section with a date. The repository also describes an optional visual dashboard delivered as an HTML file, which can be copied into the working directory the first time the task system is used. That dashboard reads and writes the same TASKS.md file, auto-saves changes, detects edits made externally via the command line, and supports drag-and-drop reordering of tasks and sections.

Practical value, limits, and safety

The main appeal of this approach is its simplicity: because everything lives in a single plain-text Markdown file, it can be opened with any ordinary editor, checked into version control, or processed by other tools without needing a separate database or cloud service. At the same time, that also means the skill offers no reminders, notifications, or multi-device, multi-user synchronization; it is designed as a lightweight, local tool rather than a full project management system. When extracting tasks from meeting notes or conversations, the stated convention requires the assistant to ask before adding entries rather than adding them automatically, which prevents unwanted or misattributed tasks from cluttering the list. Anyone who records sensitive details, such as client names or internal project information, inside TASKS.md should keep in mind that the file is treated like any other local text file and should be protected with the same access controls and backup practices as the rest of the project directory.

Free
Provider
Anthropic
License
Apache-2.0
Last reviewed
09.09.2026

Repository and documentation

Categories

Compatible with

Claude Code