Planning and Task Management skill 02 of 13
Task System Setup (Agent-Maintained Task List) SKILL.md
The Task System Setup skill makes Claude, Codex or another AI agent set up a plain-text task system you share: a TASKS.md file with Today, This week, Waiting for, Later, Projects and Done, and a MEMORY.md glossary that decodes your shorthand, people and project codenames. It rewrites vague items as next actions, agrees update rules and runs morning, after-meeting and weekly syncs.
task-system-setup/SKILL.md, version 1.0.0, 1,524 words.
By Afeez Adebayo, Marketing, SEO & Growth Expert with 10+ years of experience.
Created 10 October 2026. Updated 10 October 2026. Free to use under CC BY 4.0.
AI agents are good at keeping lists tidy, but only if they know where the list lives, how it is structured and what the user's shorthand means. Without that, every session starts from zero: the agent asks the same questions, misreads "send PSR to Todd re Phoenix", and tasks drift between chats. Anthropic's open-source productivity plugin tackles this with a plain task file and a working-memory file that decodes the user's own terms, people and projects. [Reported, anthropics/knowledge-work-plugins]
This skill sets up and maintains a simple, plain-text task system that a person and their AI agent share: a TASKS.md file with clear sections, a glossary that decodes shorthand, rules for adding, updating and closing tasks, and routines for syncing tasks from meetings, email and chat.
Operating principles
- Plain text, one file. A Markdown task file works in any agent and any editor.
- Next actions, not topics. Every open task starts with a verb and is small enough to do.
- Decode once, remember always. Shorthand, nicknames and codenames go into a glossary the agent reads first.
- The user owns priorities. The agent proposes; the user decides what matters.
- Keep it light. If maintenance takes more than a few minutes a day, simplify.
The files
TASKS.md
# Tasks
## Today
- [ ] Draft Q4 budget email to Sam (due Tue) #finance
## This week
- [ ] Review launch brief from Maya #launch
## Waiting for
- [ ] Contract signature from Acme (asked 3 Oct) #legal
## Later
- [ ] Research new CRM options #ops
## Projects
- Launch: next action "Book review with design" (see projects/launch.md)
## Done (this week)
- [x] Send invoice to Northwind (2 Oct)MEMORY.md (working memory and glossary)
## People
Sam: finance lead, approves budgets
## Terms and codenames
PSR: product status report, weekly, sent Fridays
Phoenix: the website redesign project
## Preferences
Due dates in day-month format; British EnglishLarger projects can have their own file under projects/ with goal, milestones and next actions.
When to use, and when not to
Use it when setting up an AI-assisted task list, moving from scattered lists to one file, teaching an agent the user's shorthand, or syncing tasks from notes, email and chat.
Don't use it when a team must use a shared project tool such as Asana, Jira or Linear; in that case mirror only personal tasks, or connect the agent to the shared tool.
Raw SKILL.md file
This is the exact file agents read, including the credit header and licence notice. Save it as task-system-setup/SKILL.md.
--- name: task-system-setup description: "Set up a plain-text task system a person and their AI agent share: a TASKS.md file with Today, This week, Waiting for, Later, Projects and Done, a MEMORY.md glossary that decodes shorthand, people and codenames, next-action rewrites, update rules and morning, after-meeting and weekly sync routines. Use whenever someone wants an AI agent to manage their to-do list, set up TASKS.md, consolidate scattered tasks, teach an agent their shorthand or sync tasks from meetings, email and chat." license: CC-BY-4.0 metadata: author: Afeez Adebayo author_url: https://afeezadebayo.com author_linkedin: https://www.linkedin.com/in/afeez-adebayo/ author_x: https://x.com/za_icon source: https://afeezadebayo.com/skills-md/productivity/planning-tasks/task-system-setup/ category: productivity subcategory: planning-tasks version: "1.0.0" created: "2026-10-10" updated: "2026-10-12" --- # Task System Setup (Agent-Maintained Task List) AI agents are good at keeping lists tidy, but only if they know where the list lives, how it is structured and what the user's shorthand means. Without that, every session starts from zero: the agent asks the same questions, misreads "send PSR to Todd re Phoenix", and tasks drift between chats. Anthropic's open-source productivity plugin tackles this with a plain task file and a working-memory file that decodes the user's own terms, people and projects. [Reported, anthropics/knowledge-work-plugins] This skill sets up and maintains a simple, plain-text task system that a person and their AI agent share: a `TASKS.md` file with clear sections, a glossary that decodes shorthand, rules for adding, updating and closing tasks, and routines for syncing tasks from meetings, email and chat. ## Operating principles 1. **Plain text, one file.** A Markdown task file works in any agent and any editor. 2. **Next actions, not topics.** Every open task starts with a verb and is small enough to do. 3. **Decode once, remember always.** Shorthand, nicknames and codenames go into a glossary the agent reads first. 4. **The user owns priorities.** The agent proposes; the user decides what matters. 5. **Keep it light.** If maintenance takes more than a few minutes a day, simplify. ## The files **`TASKS.md`** ``` # Tasks ## Today - [ ] Draft Q4 budget email to Sam (due Tue) #finance ## This week - [ ] Review launch brief from Maya #launch ## Waiting for - [ ] Contract signature from Acme (asked 3 Oct) #legal ## Later - [ ] Research new CRM options #ops ## Projects - Launch: next action "Book review with design" (see projects/launch.md) ## Done (this week) - [x] Send invoice to Northwind (2 Oct) ``` **`MEMORY.md` (working memory and glossary)** ``` ## People Sam: finance lead, approves budgets ## Terms and codenames PSR: product status report, weekly, sent Fridays Phoenix: the website redesign project ## Preferences Due dates in day-month format; British English ``` Larger projects can have their own file under `projects/` with goal, milestones and next actions. ## When to use, and when not to Use it when setting up an AI-assisted task list, moving from scattered lists to one file, teaching an agent the user's shorthand, or syncing tasks from notes, email and chat. Don't use it when a team must use a shared project tool such as Asana, Jira or Linear; in that case mirror only personal tasks, or connect the agent to the shared tool. ## Workflow ### Step 1. Find the existing tasks Ask where tasks live now. Import them into `TASKS.md`, grouped by section. ### Step 2. Decode shorthand For each task, find abbreviations, nicknames, project codenames and unclear references. Ask only about terms not yet decoded, and add each answer to `MEMORY.md`. ### Step 3. Rewrite as next actions Turn vague items into verb-led actions with a clear finish: "Website" becomes "Send homepage copy to design for review". ### Step 4. Add useful metadata Due dates only where real, a person for waiting items, tags for areas, and links to source messages. ### Step 5. Agree the rules | Event | Agent action | |---|---| | New request in a message or meeting | Add to Today, This week or Later with source and date | | Task done | Tick it and move to Done with date | | Someone else owes an action | Move to Waiting for with person and date asked | | Task untouched for 3 weeks | Flag at the weekly review: do, schedule, delegate or delete | | Unknown term | Ask once, then add to the glossary | ### Step 6. Set the sync routines - **Morning:** read `TASKS.md`, suggest Today's top three, flag overdue items. - **After meetings:** add actions from notes, with owners and dates. - **Weekly:** clean Done, review Waiting for, archive stale items, update projects. ### Step 7. Keep memory current Add new people, terms and projects as they appear. Keep `MEMORY.md` short, ideally under 100 lines, with detail in separate files. ## Writing good next actions | Vague | Next action | |---|---| | Budget | Email Sam the three cost options for Q4 budget | | Website | Send homepage copy to design for review | | Hiring | Write the job description draft for the analyst role | | Taxes | Collect last year's invoices into one folder | A good next action names a verb, an object and, where it helps, a person and a finish line. If the action takes more than about two hours, it is probably a project; give it a file and a first step. ## Tags and areas Keep tags few and stable, such as #finance, #launch, #hiring and #personal. Areas help the weekly review spot neglected parts of life or work. Avoid tags for priority; the Today and This week sections already do that job. ## Privacy Task files often contain names, client details and personal matters. Store them in a private folder or repository, avoid sharing the glossary publicly, and keep anything sensitive out of files that sync to shared tools. ## Working with calendars Dated tasks and appointments belong in the calendar; the task file holds actions. When a task gets a time block, note the date beside it so the morning plan can see what is already scheduled. ## Recurring tasks List recurring work, such as a weekly report or monthly invoices, in a short Recurring section with the rhythm and next due date. Each morning the agent copies due items into Today, so routine work never depends on memory. ## Setting up in different agents In Claude Code or Codex, keep `TASKS.md` and `MEMORY.md` in the project or home folder the agent opens. In chat apps without file access, paste the files at the start of a session or store them in a connected drive. The structure stays the same everywhere, which keeps the system portable. ## When to simplify If the files grow faster than the user can review them, cut sections, merge tags and archive old projects. A smaller list that is trusted beats a complete one that is ignored. ## Edge cases and judgement calls | Situation | What to do | |---|---| | Hundreds of old tasks | Move anything older than 60 days to an archive; review later | | Sensitive tasks or names | Keep them out of shared files; use initials if needed | | Several agents or devices | One file in a synced folder or repository | | Team tool required | Mirror personal tasks only; link to the team tool | | User dislikes Markdown | Use the same structure in their notes app | ## Handoffs to related skills - Audit first: `productivity-system-audit` - Capturing from many sources: `task-capture-clarify` - Choosing Today's tasks: `prioritisation`, `daily-planning` - Weekly clean-up: `weekly-review` - Notes and knowledge files: the Knowledge Management skills ## Worked example An operations lead kept tasks in chat threads and sticky notes. The agent imported 46 items into `TASKS.md`, decoded 9 terms (including two project codenames and a recurring report acronym) into `MEMORY.md`, rewrote 18 vague items as next actions, moved 7 items to Waiting for with names and dates, and archived 11 stale ones. Each morning the agent proposes three priorities; after each meeting it adds actions from the notes. The lead reports spending five minutes a day on the list instead of chasing messages. ## Output format ``` 1. TASKS.md (complete, ready to save) 2. MEMORY.md additions (people, terms, preferences) 3. Rules the agent will follow 4. Daily and weekly sync routines 5. Questions for terms still unknown ``` ## Common mistakes - Topics instead of next actions. - Tasks in several files and apps. - The agent changing priorities without asking. - A glossary that grows into a diary. - No weekly clean-up. ## Self-check before you answer - Does every open task start with a verb? - Is all shorthand decoded or asked about? - Are the update rules clear? - Is the system light enough to keep up daily? --- Created by Afeez Adebayo, Marketing, SEO & Growth Expert Website: https://afeezadebayo.com LinkedIn: https://www.linkedin.com/in/afeez-adebayo/ X: https://x.com/za_icon Source: https://afeezadebayo.com/skills-md/productivity/planning-tasks/task-system-setup/ Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). You may copy, share and adapt this skill, including for commercial use, as long as you credit Afeez Adebayo and link to the source URL above. Keep this notice in any copy or adapted version.
Workflow
Step 1. Find the existing tasks
Ask where tasks live now. Import them into TASKS.md, grouped by section.
Step 2. Decode shorthand
For each task, find abbreviations, nicknames, project codenames and unclear references. Ask only about terms not yet decoded, and add each answer to MEMORY.md.
Step 3. Rewrite as next actions
Turn vague items into verb-led actions with a clear finish: "Website" becomes "Send homepage copy to design for review".
Step 4. Add useful metadata
Due dates only where real, a person for waiting items, tags for areas, and links to source messages.
Step 5. Agree the rules
| Event | Agent action |
|---|---|
| New request in a message or meeting | Add to Today, This week or Later with source and date |
| Task done | Tick it and move to Done with date |
| Someone else owes an action | Move to Waiting for with person and date asked |
| Task untouched for 3 weeks | Flag at the weekly review: do, schedule, delegate or delete |
| Unknown term | Ask once, then add to the glossary |
Step 6. Set the sync routines
- Morning: read
TASKS.md, suggest Today's top three, flag overdue items. - After meetings: add actions from notes, with owners and dates.
- Weekly: clean Done, review Waiting for, archive stale items, update projects.
Step 7. Keep memory current
Add new people, terms and projects as they appear. Keep MEMORY.md short, ideally under 100 lines, with detail in separate files.
Writing good next actions
| Vague | Next action |
|---|---|
| Budget | Email Sam the three cost options for Q4 budget |
| Website | Send homepage copy to design for review |
| Hiring | Write the job description draft for the analyst role |
| Taxes | Collect last year's invoices into one folder |
A good next action names a verb, an object and, where it helps, a person and a finish line. If the action takes more than about two hours, it is probably a project; give it a file and a first step.
Tags and areas
Keep tags few and stable, such as #finance, #launch, #hiring and #personal. Areas help the weekly review spot neglected parts of life or work. Avoid tags for priority; the Today and This week sections already do that job.
Privacy
Task files often contain names, client details and personal matters. Store them in a private folder or repository, avoid sharing the glossary publicly, and keep anything sensitive out of files that sync to shared tools.
Working with calendars
Dated tasks and appointments belong in the calendar; the task file holds actions. When a task gets a time block, note the date beside it so the morning plan can see what is already scheduled.
Recurring tasks
List recurring work, such as a weekly report or monthly invoices, in a short Recurring section with the rhythm and next due date. Each morning the agent copies due items into Today, so routine work never depends on memory.
Setting up in different agents
In Claude Code or Codex, keep TASKS.md and MEMORY.md in the project or home folder the agent opens. In chat apps without file access, paste the files at the start of a session or store them in a connected drive. The structure stays the same everywhere, which keeps the system portable.
When to simplify
If the files grow faster than the user can review them, cut sections, merge tags and archive old projects. A smaller list that is trusted beats a complete one that is ignored.
Edge cases and judgement calls
| Situation | What to do |
|---|---|
| Hundreds of old tasks | Move anything older than 60 days to an archive; review later |
| Sensitive tasks or names | Keep them out of shared files; use initials if needed |
| Several agents or devices | One file in a synced folder or repository |
| Team tool required | Mirror personal tasks only; link to the team tool |
| User dislikes Markdown | Use the same structure in their notes app |
Handoffs to related skills
- Audit first:
productivity-system-audit - Capturing from many sources:
task-capture-clarify - Choosing Today's tasks:
prioritisation,daily-planning - Weekly clean-up:
weekly-review - Notes and knowledge files: the Knowledge Management skills
Worked example
An operations lead kept tasks in chat threads and sticky notes. The agent imported 46 items into TASKS.md, decoded 9 terms (including two project codenames and a recurring report acronym) into MEMORY.md, rewrote 18 vague items as next actions, moved 7 items to Waiting for with names and dates, and archived 11 stale ones. Each morning the agent proposes three priorities; after each meeting it adds actions from the notes. The lead reports spending five minutes a day on the list instead of chasing messages.
Output format
1. TASKS.md (complete, ready to save)
2. MEMORY.md additions (people, terms, preferences)
3. Rules the agent will follow
4. Daily and weekly sync routines
5. Questions for terms still unknownCommon mistakes
- Topics instead of next actions.
- Tasks in several files and apps.
- The agent changing priorities without asking.
- A glossary that grows into a diary.
- No weekly clean-up.
Self-check before you answer
- Does every open task start with a verb?
- Is all shorthand decoded or asked about?
- Are the update rules clear?
- Is the system light enough to keep up daily?
On this page
- When to use this skill, and when not to
- Example prompts
- Example output
- What the agent needs from you
- How this skill hands off to related skills
- How to run Task System Setup step by step
- Situations this skill handles
- What makes this skill different
- Practical tips from real projects
- Frequently asked questions
- Changelog
When to use this skill, and when not to
Use it when you want an AI agent to manage your to-do list, consolidate scattered tasks, teach the agent your shorthand or sync tasks from meetings, email and chat.
Without a shared file and glossary, every agent session starts from zero and misreads notes such as send PSR to Todd about Phoenix. Anthropic's open-source productivity plugin uses the same idea of a task file plus working memory.
Don’t use it when your team must use a shared project tool; mirror only personal tasks or connect the agent to that tool.
Example prompts
- “Set up a TASKS.md my agent can maintain.”
- “Import my tasks from these notes and chats.”
- “Learn my abbreviations and project names.”
- “Add actions from this meeting to my list.”
- “What should I focus on today from my task file?”
Example output
The skill returns a fixed structure. A shortened example for an imaginary site:
Imported: 46 items into TASKS.md Glossary: 9 terms decoded into MEMORY.md (2 codenames, 1 report acronym) Rewritten as next actions: 18 Waiting for: 7 with names and dates Archived as stale: 11 Routine: morning top three, after-meeting sync, weekly clean-up
Every open task starts with a verb, every unknown term is asked about once and remembered, and you keep control of priorities.
What the agent needs from you
The skill asks where your tasks live now, the terms, people and projects you mention often, how you want dates and areas written and whether the files will sit in a project folder, a synced folder or a connected drive.
How this skill hands off to related skills
Task System Setup connects to the Productivity System Audit, Task Capture and Clarify, Prioritisation and the Weekly Review, plus the Knowledge Management skills planned in the Productivity category.
How to run Task System Setup step by step
- Install the skill as
task-system-setup/SKILL.md. - Gather your current tasks.
- Answer questions about unknown terms.
- Review the next actions.
- Agree update rules.
- Run the daily and weekly syncs.
Situations this skill handles
Task lists vary. The skill handles these:
| Situation | What the skill does |
|---|---|
| Hundreds of old tasks | Older than 60 days archived |
| Sensitive names | Kept out of shared files |
| Several devices | One synced file |
| Team tool required | Personal tasks mirrored |
| Dislikes Markdown | Same structure in notes app |
| Recurring work | Recurring section with due dates |
What makes this skill different
Most AI to-do tools forget everything between sessions. This skill gives the agent a lasting task file and a glossary of your own language, so it understands your notes and keeps the list current without taking over your priorities.
Because it is plain text, it works in every agent you use.
Practical tips from real projects
- Plain text, one file.
- Next actions, not topics.
- Decode once, remember always.
- The user owns priorities.
A worked example: an operations lead kept tasks in chat threads and sticky notes. The agent imported 46 items into TASKS.md, decoded nine terms, including two project codenames and a recurring report acronym, into MEMORY.md, rewrote 18 vague items as next actions, moved seven to Waiting for with names and dates and archived 11 stale ones. Each morning the agent proposes three priorities; after each meeting it adds actions from the notes. The lead now spends about five minutes a day on the list instead of chasing messages.
The task file has six sections: Today, This week, Waiting for with the person and date asked, Later, Projects with a next action each and a link to a project file, and Done for the current week. The glossary lists people and their roles, terms and codenames with their meanings and preferences such as date formats and British English.
Update rules keep the file reliable: new requests from messages or meetings are added with source and date; done tasks are ticked and moved to Done; actions owed by others move to Waiting for; tasks untouched for three weeks are raised at the weekly review to do, schedule, delegate or delete; and unknown terms are asked about once and added to the glossary.
Good next actions name a verb, an object and, where it helps, a person and a finish line. Budget becomes email Sam the three cost options for the Q4 budget, and website becomes send homepage copy to design for review. Anything over about two hours becomes a project with its own file and a first step.
Tags stay few and stable, such as finance, launch, hiring and personal, and priority is shown by section rather than tags. Recurring work sits in its own section with a rhythm and next due date, and the agent copies due items into Today each morning. Dated appointments belong in the calendar, while the task file holds actions. Files stay private because they often contain client and personal details, and in apps without file access they can be pasted in or stored in a connected drive with the same structure.
Daily routines keep the system light: in the morning the agent reads the file, suggests the top three tasks and flags overdue items; after meetings it adds actions with owners and dates; and each week it clears Done, reviews waiting items and archives stale tasks. If maintenance takes more than a few minutes a day, sections and tags are cut until it does not.
When files grow faster than the user can review them, sections are cut, tags are merged and old projects are archived, because a smaller list that is trusted beats a complete one that is ignored. Hundreds of inherited tasks can be moved to a dated archive and reviewed later.
The same two files can serve several agents at once, so the user gets the same answers about their tasks in Claude Code, Codex or a chat app.
Frequently asked questions
What is TASKS.md?
A plain Markdown file with sections for Today, This week, Waiting for, Later, Projects and Done that you and your AI agent both read and update.
What goes in MEMORY.md?
People, terms, project codenames and preferences the agent needs to understand your tasks, kept under about 100 lines.
Why plain text instead of an app?
It works in any agent and editor, is easy to back up and keeps the system portable.
Will the agent change my priorities?
No. It proposes; you decide what matters.
How does it handle old tasks?
Items untouched for three weeks are flagged at the weekly review, and anything older than 60 days can be archived.
Can it work with a team tool such as Jira or Asana?
Yes, by mirroring only personal tasks and linking to the team tool.
Changelog
- 1.0.0, 10 October 2026: first release.
Use, adapt and share this skill under CC BY 4.0: credit Afeez Adebayo and link to this page.
