Morning, midday, and evening summaries
Prepare time-specific task summaries for the conversation or a separately authorized push channel, without treating a cached list as current truth.
Task planner Skill
Give a compatible AI workspace a reusable way to capture tasks, review the week, triage an inbox, and propose schedule changes while WeeklyPlanner remains the source of truth.
The current V1 Skill routes explicit WeeklyPlanner requests into focused workflows instead of applying one oversized prompt to every situation.
Prepare time-specific task summaries for the conversation or a separately authorized push channel, without treating a cached list as current truth.
Review unscheduled tasks, separate unclear items from actionable work, and propose dates or groups before changing the plan.
Use day-review and planning context where available, preserve coverage limits, and distinguish historical commitments from a current snapshot.
Preview task moves, ask for explicit confirmation, apply approved patches atomically, and read the affected tasks back after the write.
Create tasks, list today or this week, find overdue work, mark numbered results complete, move tasks, delete tasks, and inspect groups.
Summarize aggregate planning context and task patterns without exposing raw event rows or inventing a fulfillment rate when coverage is incomplete.
A real plan, not a mock integration
Create a revocable key in WeeklyPlanner, connect a compatible agent, and let the workflow read or update the same plan you can inspect in the app.

The Skill, connector, and planner have different jobs. Keeping those layers separate makes the workflow easier to inspect and safer to update.
It defines when to use WeeklyPlanner, which workflow to run, how to interpret task state, and when the agent must preview or confirm a change.
A registered WeeklyPlanner connector maps logical operations such as list_tasks and update_task to scoped product access. The Skill never bypasses that boundary.
Tasks, groups, dates, time blocks, and review data stay in WeeklyPlanner. The agent reads back affected tasks instead of maintaining a separate invisible list.
The current Skill is a local directory containing SKILL.md, workflows, and references. There is no universal installer across AI clients, so use the local Skill import method documented by your compatible client and keep the directory intact.
Import or copy task-planner through the client-supported local Skill mechanism. Do not copy only SKILL.md; its workflows and references are part of the operating contract.
Ask the agent to use task-planner for a WeeklyPlanner request. The Skill discovers session capabilities at runtime instead of assuming a connector is available.
If the session cannot expose WeeklyPlanner operations, configure or register the connector through your platform's approved setup flow. Installing the Skill does not install, authenticate, or restart a connector.
List tasks for a specific day or week without writes. Only after the result is correct should you authorize one clearly scoped create or update and verify it by task ID.
Compatibility means the runtime can load a local Skill and expose equivalent WeeklyPlanner logical operations. It does not mean every AI product or chat client is automatically supported.
The Skill can run where Codex exposes local Skills and the current session can discover the registered WeeklyPlanner connector operations it needs.
The same workflow directory can be used in Claude environments that support local Skills and expose an equivalent connector at runtime.
Treat support as unverified until the runtime can load SKILL.md, preserve its referenced files, and map the required logical operations. A generic prompt import is not equivalent.
The Skill orchestrates the subset of task, group, settings, review, and planning operations required by a workflow. It does not promise to call every MCP tool, and it does not expand connector permissions.
A request for a specified day, week, or filter authorizes that read. The agent should not query unrelated account data merely to test the connection.
Create, update, complete, delete, and batch actions require a current instruction that identifies the intended operation. Writes are followed by a task-level read-back.
The agent previews proposed dates or time blocks, waits for explicit confirmation, uses conflict-aware batch patches, and does not batch-move recurring children.
Connector registration, credentials, push-channel login, test messages, and scheduled jobs each require separate authorization. None happens merely because the Skill was installed.