--- title: "Notion Custom Agents: Automate Your Task Checklists" description: "Build a Notion Custom Agent that turns task descriptions into three-step checklists. Set permissions, test database triggers, and inspect saved results." canonical: "https://adamtheautomator.com/notion-custom-agents-task-checklists/" --- # Notion Custom Agents: Automate Your Task Checklists > Build a Notion Custom Agent that turns task descriptions into three-step checklists. Set permissions, test database triggers, and inspect saved results. Source: https://adamtheautomator.com/notion-custom-agents-task-checklists/ --- ATA Learning Tap to hide [ ATA Learning ](/) * [Home](/) * [Tutorials](/tutorials/) * [Instructors](/author/) * [Advertising](/advertising/) * [Recommended Resources](/resources/) * [About Adam](/about-adam/) Search for: * [](https://twitter.com/adbertram) * [](https://github.com/Adam-the-Automator) * [](https://www.linkedin.com/company/adam-the-automator-llc) * [](/feed/) ![Notion Custom Agents: Automate Your Task Checklists](https://adamtheautomator.com/wp-content/uploads/publisher/3f25d9c85b2b81a897a0e1623e958463/cb5678abd6a8724484f6f9f958245d4db7e58d587e9f2a52d22e92432f546c7d.webp) # Notion Custom Agents: Automate Your Task Checklists [![](https://secure.gravatar.com/avatar/d0b9d42e21e5622713f8b693aa5c0f9244d5f7dd200ed29b8398f52dee5de337?s=192&d=mm&r=g)Adam Bertram](https://adamtheautomator.com/author/adam-bertram/)9 October 20268 min. read Categories: [AI](/category/ai/) Tags:[Notion](/tag/notion/)[AI](/tag/ai/)[Automation](/tag/automation/) Table of Contents * [Prerequisites](#prerequisites) * [Prepare a Task Database](#prepare-a-task-database) * [Create a Notion Custom Agent and Limit Its Access](#create-a-notion-custom-agent-and-limit-its-access) * [Give the Agent One Precise Job](#give-the-agent-one-precise-job) * [Paste the Instructions](#paste-the-instructions) * [Check the Boundaries](#check-the-boundaries) * [Test One Prepared Task in Chat](#test-one-prepared-task-in-chat) * [Verify the Saved Checklist](#verify-the-saved-checklist) * [Run Automatically When a Task Is Ready](#run-automatically-when-a-task-is-ready) * [Configure the Ready Condition](#configure-the-ready-condition) * [Check the Automatic Result](#check-the-automatic-result) * [Verify Reruns and Diagnose Failures](#verify-reruns-and-diagnose-failures) * [Check Credit Use and Stop the Demo](#check-credit-use-and-stop-the-demo) Turn a task description into three actions you can hand to someone. Start with a request to collect last quarter’s campaign spending and identify the best performers. Include a short presentation as the deliverable. Keep the description beside the resulting checklist so you can check whether the agent preserved what you asked for. Build your first [Notion Custom Agent](https://www.notion.com/help/custom-agents) around that single job in a private task database. Judge manual requests and automatic database runs by the checklists they save. Reject a checklist that invents campaign results, even if the run reports success. ## Prerequisites Keep this first build in a private test area with synthetic tasks. You’ll need: * A [Business or Enterprise workspace](https://www.notion.com/help/custom-agents) with permission to create Notion Custom Agents. A restricted creation policy can prevent you from building one even on an eligible plan. * Available Notion credits so the agent can execute your tests. Check the workspace’s [credit settings](https://www.notion.com/help/buy-and-track-notion-credits-for-custom-agents) before you start; a depleted balance pauses agent execution. * Full Access to your demonstration database so you can grant the agent permission to edit it. [Notion’s permission requirements](https://www.notion.com/help/custom-agents-sharing-and-permissions) distinguish granting view access from granting broader access. * Familiarity with creating a Notion database and adding properties. You’ll use text and checkbox properties, with no code or external integration required. Your Custom Agent keeps its instructions between runs and has its own resource access. The instructions define the checklist it writes into the task database. Checking **Ready** triggers processing. Check database permissions separately when troubleshooting; clearer wording cannot repair missing edit access. ## Prepare a Task Database Create a table database named **Agent Demo Tasks** in your private test area. Give it these properties, using Notion’s [database property types](https://www.notion.com/help/database-properties): | Property | Type | Purpose | | --- | --- | --- | | Name | Title | Identify the task | | Task Description | Text | Hold the request to break down | | Ready | Checkbox | Signal that you’ve finished entering the request | | Action Checklist | Text | Store the agent’s three numbered action lines | The output is a text checklist. It won’t give you individually clickable completion boxes. Keeping the output in one property makes it easy to compare before and after a run without searching a page for appended content. Add a row named **Prepare Marketing Review**. Enter this in **Task Description**: > Gather the advertising spend for July through September and identify the three best-performing campaigns. Prepare a five-slide presentation for the marketing team. Leave **Ready** unchecked and **Action Checklist** empty. Finish the input before marking it ready. You can create the row and fill in its description without starting a run. * Keep the title and description visible beside the output when checking results. Use the completed sample as the input for your first agent test. ## Create a Notion Custom Agent and Limit Its Access Open **Chat** in the Notion sidebar, hover over **Agents**, choose **New agent**, then select **Create blank**. Name your agent **Task Checklist Builder**. The workflow was tested in Notion’s web app on October 7, 2026 (America/Chicago). No build or version identifier was exposed. Open the agent’s **Settings** and find **Tools and access**. Keep workspace-wide Notion access set to **None**. Add only **Agent Demo Tasks**, granting **Can edit content** so the agent can write its output. Keep web access off and leave external tools disconnected for this exercise. In _Custom Agents sharing & permissions_, Notion states: “There are no inherited permissions from anyone”. Your ability to edit the database doesn’t automatically become the agent’s ability to edit it. [Check the agent’s resource access explicitly](https://www.notion.com/help/custom-agents-sharing-and-permissions). The access configuration should show the demonstration database as the resource the agent can edit, as shown below. Screenshots use the isolated test names **atablog-5414db18de-db** and **atablog-5414db18de-agent** in place of **Agent Demo Tasks** and **Task Checklist Builder**. ![Agent database access](https://adamtheautomator.com/wp-content/uploads/publisher/e9babb3461654d20effde07f6392305cf4e255f3094675d19c374c5fef03dac4.png) * * * _**Warning: Sharing an agent can let other people retrieve information through its resource access. Keep this demonstration private, and review accessible content before sharing an agent with teammates.**_ * * * Confirm the correct database is selected before adding instructions. Another database with a similar name would give you a plausible-looking configuration that writes to the wrong place. ## Give the Agent One Precise Job Your instructions need to define what the agent may change and when it should stop. Notion’s [agent optimization guidance](https://www.notion.com/help/best-practices-for-creating-and-optimizing-a-custom-agent) recommends a focused job and explicit completion criteria. For this example, a completed checklist is enough; the agent must never perform the marketing tasks it describes. ### Paste the Instructions Add the following to **Instructions**. If you rename the database or properties, update those names here too. ```text You turn one task into a concise action checklist. Use only the Agent Demo Tasks database. Process only the task provided by the trigger or explicitly linked in the current chat. If no single task is identified, ask for its link and stop. Read Name, Task Description, Ready, and Action Checklist. If Ready is unchecked, stop without changing the task. If Task Description is blank, stop without inventing a checklist. If Action Checklist already contains any text, leave it unchanged and stop. Never append a second checklist. For an eligible task, write exactly three numbered lines to Action Checklist on that same task. Start each line with an action verb. Use only information in Task Description. Preserve stated quantities and constraints. Do not invent people, dates, links, campaign results, or completed work. Do not change Name, Task Description, Ready, or page content. Do not execute the actions in the checklist. Do not create tasks, send messages, or search other sources. Read back Action Checklist to check the saved result. Finish when the three lines are stored, or explain why you made no change. ``` ### Check the Boundaries The Task Checklist Builder instructions require testing and do not guarantee enforcement. The existing-output check protects your review edits from routine reruns, but natural-language instructions aren’t a transaction lock for simultaneous requests. | Input state | Intended behavior | | --- | --- | | Ready unchecked | Leave the task alone | | Description blank | Write nothing | | Checklist already present | Preserve existing text | | Ready checked, description present, checklist empty | Write three numbered lines | Keep the model on **Auto** for the first comparison. Click **Save** in agent Settings, with no automatic trigger configured yet. Wait until **Save** is disabled and the unsaved-edit notice is gone. Reopen **Instructions** and confirm the full block remains, including the final readback rule. The first run should be one you can inspect directly. ## Test One Prepared Task in Chat Check **Ready** on **Prepare Marketing Review**. On the agent, choose **Share**, then **Copy link**. Paste that link into a blank content block on a private scratch page and choose **Embed**. Enter the following request in the embedded agent, including a direct link to the database row, then choose **Submit AI message**: > Process this task and save exactly three numbered lines to its Action Checklist property: \[paste the task’s Notion link here\]. Link the row directly to distinguish it from tasks with the same title. Notion documents [agent chat and embedding](https://www.notion.com/help/custom-agents). Use your synthetic row because the manual run can write real data. The configuration assistant can discuss or change the agent’s setup. Its response, or a **Test run, no side effects** entry, doesn’t establish that the agent saved a checklist. The embedded-agent route was verified before adding any database trigger. Judge success by reopening the task, regardless of the chat’s textbox label. ### Verify the Saved Checklist Open the row independently after the agent finishes. The live manual-first test saved three numbered lines while preserving the requested period and quantities. Use this illustrative example to compare your result: ```text 1. Gather advertising spend data for July through September. 2. Identify the three best-performing campaigns. 3. Prepare a five-slide presentation for the marketing team. ``` The exact wording can differ. Check that **Action Checklist** contains three lines. The text must preserve the requested period and quantities without claiming the work has already happened. Compare **Name**, **Task Description**, and **Ready** with their original values; all should remain unchanged. The completed row should let you compare the saved input with the checklist, as shown below. ![Saved task checklist](https://adamtheautomator.com/wp-content/uploads/publisher/37d0640ecf0956d0d3d579fa164b67645672b404bafba10388dbc80fdfc0c92b.png) If the output appears only in Chat, the test has failed. Fix the write instruction or database access before adding automatic runs. ## Run Automatically When a Task Is Ready After the manual run saves your checklist, test whether the database event identifies a new task without a Chat request. Notion supports [database property-change triggers and conditions](https://www.notion.com/help/custom-agents). ### Configure the Ready Condition In **Settings**, choose **Add trigger**, then **Notion: Property updated in database**. Select **Agent Demo Tasks**, choose **Add property**, and select **Ready**. Both **Unchecked** and **Checked** initially have checkmarks. Deselect **Unchecked**, leaving **Checked** selected, then choose **Done**. Confirm the condition chip reads **Ready set to checked**. If you click **Checked** while both values are selected, you remove the value you need; the resulting trigger watches for an unchecked box. Leave **Trigger when page content edited** unchecked. Choose **Add trigger** in the dialog, then click **Save** in agent Settings. Wait for **Save** to become disabled with no unsaved-edit notice. Avoid an unrestricted property-change event that also fires when the agent writes **Action Checklist**. Inspect the selected property and checked-value condition before testing, as shown below. ![Ready trigger condition](https://adamtheautomator.com/wp-content/uploads/publisher/5afc4911efe5da8b83bc1c81b2a70abb3e94bce5f472f3612b896c097f7955b3.png) ### Check the Automatic Result Create a second row called **Prepare Workshop Materials**. Enter this description while **Ready** is unchecked: > Collect the workshop exercises and check that every exercise has a solution. Put the materials in a folder for attendees. Leave **Action Checklist** empty, then check **Ready**. Don’t send a Chat message. Inspect the row for a saved checklist and find the matching automatic run under **Recent chats** on the agent home page. Use **Refresh activity** if needed. Record how long your test takes; don’t assume immediate completion or create repeated requests while waiting. In the live test, the saved workshop checklist was visible within 30 seconds of checking **Ready**. Its run entry reported **Worked for 2s**. The elapsed time to observe a saved row differs from the reported execution duration; neither measurement guarantees your next run’s timing. ## Verify Reruns and Diagnose Failures Open the matching entry under **Recent chats** to inspect its trigger and actions. Follow its **View page** link to confirm the triggering task. Read any reported errors before retrying. Notion’s [run-debugging guidance](https://www.notion.com/help/custom-agents) describes inspecting run activity. Match the run to your workshop task, rather than accepting an unrelated successful entry. Test the boundaries before using the agent on a real request: | Check | Pass condition | | --- | --- | | Process the completed task again through Chat | Existing checklist stays unchanged | | Ready task with a blank description | No checklist invented | | Prepared task with Ready unchecked | No checklist written | | Second ready task, no Chat message | Checklist saved through an automatic run | For the blank-description test, use another synthetic row. Clear that row’s **Ready** checkbox after checking its result. For the unchecked-row test, explicitly link the row in Chat so you exercise the instruction guard without relying on a trigger that shouldn’t fire. * * * _**Reality Check: The saved checklist needs your review even when the run reports success. Match it to the task, then check that it preserves the request’s constraints. Source fields must remain unchanged.**_ * * * If no automatic run appears, confirm that the trigger watches **Agent Demo Tasks** for the **Ready** change. Check that you saved the configuration. For a run that cannot write, inspect the database’s edit permission. Invented details in the checklist call for tighter instructions; repeat the same sample after editing them so you can compare results. ## Check Credit Use and Stop the Demo Open **Credits** on the agent, then **Agent’s monthly usage** to reach the workspace’s [Notion credits dashboard](https://www.notion.com/help/buy-and-track-notion-credits-for-custom-agents). Inspect usage associated with your tests. Keep the number of runs beside your observations. Don’t extrapolate a per-task price from an agent name or model choice; the amount of work varies with the request. When you’re finished, turn off the demonstration’s database trigger switch in agent Settings and click **Save**. Wait for the disabled **Save** button, then reload Settings to verify the trigger remains off. Administrators can also stop automated runs through the agent directory under **Settings**, **Notion AI**, **Agents**, as described in [Notion’s admin controls](https://www.notion.com/help/custom-agents-sharing-and-permissions). Removing database access alone doesn’t establish that automatic runs have stopped. Keep your synthetic task rows while reviewing the results. You should have saved outputs from both the manual and automatic runs. Compare the completed checklist before and after a repeat request to confirm that its text stayed unchanged. If any check fails, leave the trigger disabled while you correct the configuration. Once those checks pass, replace one synthetic request with a low-risk task you can review yourself before expanding the agent’s access or workload. Share this article [Share on X](https://twitter.com/intent/tweet?url=https%3A%2F%2Fadamtheautomator.com%2Fnotion-custom-agents-task-checklists%2F&text=Notion%20Custom%20Agents%3A%20Automate%20Your%20Task%20Checklists)[Share on Facebook](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fadamtheautomator.com%2Fnotion-custom-agents-task-checklists%2F)[Share on LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fadamtheautomator.com%2Fnotion-custom-agents-task-checklists%2F) ## Related Posts ![](https://adamtheautomator.com/wp-content/uploads/2026/06/featured_image_v2.png) ### [When AI Automates the Joy Out of Work](/ai-automation-work/) A personal look at how AI agents and automation can boost productivity while draining the satisfaction that makes hands-on technical work feel meaningful. ![](https://adamtheautomator.com/wp-content/uploads/2025/05/Untitled.jpg) ### [When a Script is Worth a Thousand AI Prompts](/when-script-worth-thousand-ai-prompts/) When traditional scripts outshine AI: a developer’s journey back to automation basics. ![](https://adamtheautomator.com/wp-content/uploads/publisher/3ec5d9c85b2b8151b299cdf1129b5930/07fca0c5d7b72581bdbccb06e9058b4f8b00f553e0bd60f4e2724cfa489a164d.png) ### [Run Ollama on Windows Server Without Exposing Your LLMs](/run-ollama-on-windows-server-without-exposing-your/) Run Ollama as a hardened Windows Server service with NSSM, keep the API on loopback, and front it with an IIS reverse proxy, TLS, and authentication. ## Categories * [IT Ops](/category/it-ops/) * [Cloud](/category/cloud/) * [DevOps](/category/devops/) * [Home Ops](/category/home-ops/) * [Information Security](/category/infosec/) * [Software Development](/category/software-development/) ## Site * [Home](/) * [Tutorials](/tutorials/) * [Instructors](/author/) * [Advertising](/advertising/) * [Recommended Resources](/resources/) * [About Adam](/about-adam/) Copyright 2026© ATA Learning | [Privacy Policy](/privacy/)