How to Install the WeeklyPlanner Task Planner Skill

What you are installing

A task planner Skill is not a second planner and it is not a single prompt. It is a reusable directory of instructions, workflows, and references that tells a compatible AI agent how to work with your real tasks.

The current WeeklyPlanner Task Planner Skill uses WeeklyPlanner as the source of truth. It can route a request into morning, midday, or evening summaries; inbox triage; a weekly review; schedule assistance; or direct task management. The Skill supplies the operating method, while a separately registered connector supplies structured task tools.

That distinction matters during installation. Copying the Skill does not authenticate an account, register a connector, create scheduled jobs, or send a message. Those actions remain separate and require their own authorization.

Before you start

Prepare three things:

1. A Codex or Claude environment that supports local Skills.
2. The complete current task-planner directory, including SKILL.md, workflows/, and references/.
3. A WeeklyPlanner account and a matching connector that the new session can discover.

Do not copy only SKILL.md. The main file deliberately routes detailed work to the referenced workflow and safety files. Removing those files turns a tested operating contract into an incomplete prompt.

The current V1 release is distributed as a local Skill directory. AI clients do not share one universal install command, so use the local Skill import or directory mechanism documented by your client. This guide describes the parts that must stay consistent across environments without inventing a one-click installer that does not exist.

Step 1: add the complete Skill directory

Import or copy the task-planner directory into the location your AI client uses for local Skills. Preserve the folder name and its internal structure.

After import, confirm that the client can discover a Skill named task-planner. Discovery proves only that the workflow files are available. It does not prove that WeeklyPlanner tools or account access are available in the current session.

If your client offers a Skill detail view, check that referenced workflows such as setup, inbox triage, weekly review, and schedule assistance are present. Do not rewrite platform-specific paths into the Skill merely to make one machine work; platform adapters are responsible for mapping equivalent capabilities.

Step 2: start a fresh session and invoke it explicitly

Open a new session after installation. Give the agent a bounded WeeklyPlanner request and name the Skill:

Use task-planner to list my unfinished WeeklyPlanner tasks for this week. Summarize them by day, but do not create, update, complete, move, or delete anything.

This request does two useful things. It gives the Skill an exact date range to read, and it makes the no-write boundary explicit. A requested day, week, or filter authorizes only that read; the agent should not query unrelated settings or statistics merely to test whether the connection works.

The Skill discovers connector operations at runtime. If the current session does not expose the logical operation it needs, the correct result is SESSION_CAPABILITY_NOT_EXPOSED. That status describes the session. It does not prove that the connector is uninstalled, the key is missing, or the service is unavailable.

Step 3: connect WeeklyPlanner separately when needed

If the Skill loads but WeeklyPlanner operations are not exposed, follow your platform's approved connector setup flow. The MCP task manager page explains the public MCP connection and scoped-key model; your client documentation remains the authority for where local connector configuration belongs.

Keep credentials out of prompts, screenshots, repositories, and the Skill directory. The Skill records environment-variable names and non-secret preferences only. A connector or approved runtime environment should receive secret values through its protected configuration mechanism.

Registration, authentication, the first read-only call, push-channel configuration, a test message, and scheduled jobs are separate steps. Authorizing one does not authorize the next.

Step 4: verify one narrow read

Run the read-only weekly request and compare the response with the visible planner. Check:

• dates and time blocks match the weekly board;
• completed and unfinished tasks are not mixed;
• unscheduled tasks remain unscheduled instead of receiving invented dates;
• no task was created, moved, completed, or deleted;
• the response does not expose credentials or raw internal event rows.

If the agent returns a historical day review, keep coverage information intact. A null fulfillment rate must remain null. Current task counts and historical start-of-day commitments answer different questions and should not be silently combined.

The WeeklyPlanner Task Planner Skill reference lists the V1 workflows and safety boundaries in one place. Use it to distinguish an expected capability gap from an unsupported promise.

Step 5: authorize one test write

After the read path is correct, test a single reversible write. For example:

Create one task named "Verify task-planner setup" for tomorrow morning, then read that task back. Do not change any other task.

The agent should create the task, capture its task ID, and read the same task back. Verify it on the weekly board before attempting batch changes.

For an update, identify one task and one field. For completion, use the numbered result from the most recent list and ask to mark that specific item complete. For deletion, name the exact test task. Broad instructions such as “clean up everything” make it difficult to prove which write was authorized.

How schedule assistance stays controlled

Multi-task scheduling uses a stricter sequence:

1. Read the current tasks and their update versions.
2. Preview proposed dates or time blocks.
3. Wait for explicit confirmation.
4. Apply the approved changes as conflict-aware patches.
5. Read every affected task back.

The Skill does not batch-move recurring child tasks. It also avoids leaving a task with a time block or time value when its date is cleared. These rules are part of the workflow, not optional wording that can be dropped for convenience.

Daily summaries and push channels are separate

Morning, midday, and evening workflows can produce summaries in the current conversation. External delivery is optional.

Feishu/Lark and Telegram require separately configured capabilities or protected runtime variables. Selecting a channel does not authorize login, recipient selection, a test message, or a scheduled job. Each external action needs explicit approval and evidence that it succeeded.

This separation prevents a common setup mistake: treating “install the Skill” as permission to create persistent background automation. Installation makes the workflow available; it does not broaden authority.

What to test before relying on it

Use this small acceptance checklist:

• the client discovers task-planner and all referenced files;
• a new session can find the required WeeklyPlanner logical operations;
• one bounded read matches the visible plan;
• one test write is followed by task-level read-back;
• multi-task scheduling previews changes before confirmation;
• missing capabilities are reported without guessing about credentials;
• no push message or scheduled job exists unless you authorized it separately.

Once these checks pass, expand one workflow at a time. Start with direct task reads and writes, then inbox triage, schedule assistance, weekly review, and only then optional external delivery.

Keep one planner as the verification layer

The long-term value of a planning Skill is not that an agent can produce a longer answer. It is that the same repeatable workflow can operate across sessions while the visible planner remains the final verification layer.

Install the complete directory, connect capabilities deliberately, prove one narrow read, and authorize the smallest useful write. For the current compatibility, workflow, and permission contract, return to the WeeklyPlanner Task Planner Skill page.