# \[Product Launch\] Chunk Tasks - fixing flaky tests **URL:** \ **Category:** Build Environment\ **Created:** [August 18, 2025, 8:37pm UTC](https://discuss.circleci.com/t/product-launch-chunk-tasks-fixing-flaky-tests/53975 "2025-08-18T20:37:08Z")\ **Posts on this page:** 5\ **Page:** 1
**Author:** ![luisejrobles](https://sea2.discourse-cdn.com/circleci/user_avatar/discuss.circleci.com/luisejrobles/32/36575_2.png) [@luisejrobles](https://discuss.circleci.com/u/luisejrobles)\ **Post date:** [August 18, 2025, 8:37pm UTC](https://discuss.circleci.com/t/product-launch-chunk-tasks-fixing-flaky-tests/53975/1 "2025-08-18T20:37:08Z")
Hello CircleCI community! Now in beta: a new agentic capability that identifies and provides fixes for flaky tests in your CircleCI projects, helping you ship quickly with confidence by reducing time spent debugging intermittent failures. > Have feedback or feature requests? Submit them on our **[Ideas board](https://circleci.canny.io/cloud-feature-requests?selectedCategory=chunk-ai)** where you can also see existing feature requests and vote on them. ## **Getting Started** ### **Prerequisites:** 1. An [Anthropic](https://console.anthropic.com/settings/keys) or [OpenAI](https://auth.openai.com/log-in)API key to enable the agent to process and generate flaky test fixes using your existing model provider. Your source code is not stored nor used for training purposes by CircleCI. 2. **Test results stored on CircleCI** : To enable CircleCI to detect test flakiness, you need to store your test results using the `store_test_results` step in your CircleCI YML configuration file. Learn more about [collecting test data here](https://circleci.com/docs/guides/test/collect-test-data/). ### Setup: 1. Navigate to the **[CircleCI web app](https://app.circleci.com/home)** → Your organization **→ Chunk Tasks** (from the left-hand navigation) **→ Get Started** 2. **Install the CircleCI GitHub App in your desired GitHub Organization** so the agent can create Pull Requests with recommended fixes. 3. Add your [Anthropic](https://console.anthropic.com/settings/keys) or [OpenAI](https://auth.openai.com/log-in) API key. 4. Select the followed project where the task should be assigned. 5. Configure your preferred 6. The agent creates an environment to run your tests by inferring setup from your repository. If you’d like more control, you can customize it with a `.circleci/cci-agent-setup.yml` file on your default branch. Learn more in our [FAQ: Unable to run verification tests](https://discuss.circleci.com/t/product-launch-agentic-capability-fixing-flaky-tests/53975/3#p-107197-unable-to-run-verification-tests-8). ## How It Works 1. When running, the agent will identify flaky tests based on the tests that are marked flaky in [CircleCI’s Test Insights](https://circleci.com/docs/guides/insights/insights-tests/) 2. The agent will generate potential fixes to the flakiness based on the **Number of solutions to try per test** setting that was configured during setup. 3. The agent will validate solutions through multiple test runs to ensure the flakiness has been removed based on the **Number of validations run per test** setting that was configured during setup. 4. The agent will open a pull request with the proposed fix The Agent tasks list in **[CircleCI web app](https://app.circleci.com/home) →Your organization → Chunk Tasks tab** will show a row per every test being analyzed. ![Chunk Tasks page](https://global.discourse-cdn.com/circleci/original/3X/e/8/e8ba3bd56e3b0649d61f6c8e5f68f0cdeb55e8d0.png) If the agent does not run into an error while analyzing & attempting to fix a given test’s flakiness, a it opens a PR with a proposed fix. ![Chunk Task](https://global.discourse-cdn.com/circleci/original/3X/9/3/9365dcd66fcb0af3d7d482dfacf162e2fbeba108.png) ![PR created by Chunk](https://global.discourse-cdn.com/circleci/original/3X/e/3/e3c9e65b05793f6b5163bba0e859c367d43cbe39.png) Each agent task will have two tabs: - **Code Diff** : the proposed code changes - **Logs** : the agent’s reasoning and analysis ![Chunk Task detail](https://global.discourse-cdn.com/circleci/original/3X/0/8/08824f94435e0db87dcc4003178b31d6f6a099f7.png) If the agent lacks confidence in the fixes or runs into an error during execution, a PR is not created, but logs and analysis remain available for review. ![Chunk Task not fixed](https://global.discourse-cdn.com/circleci/original/3X/7/2/722d149ee4d93cb2ae8471a11be771aa8a45f165.png) ![Chunk Task detail](https://global.discourse-cdn.com/circleci/original/3X/7/b/7b1d8555b731a040aabe4ac9c2eb25cbb742f5c6.png) ## Known limitations **Editing agent task configurations** Currently, there is no way to directly edit task configuration settings including post-run commands once an agent task is created. The workaround for now is to delete the Chunk task and recreate it: - - Navigate to **Organization settings \> Chunk settings \> Delete** the current agent task. - **Create** a new agent task \> customize your settings **OpenAI Zero Data Retention Compatibility Issue** If you’re using OpenAI as your model provider and see all Chunk tasks marked as “Not fixed” with “Could not diagnose a fix” messages and empty Logs tabs, the issue may be that your OpenAI account has Zero Data Retention enabled. Chunk does not yet support OpenAI accounts with Zero Data Retention. ## Ad-hoc tasks In the CircleCI web app, navigate to **Organization settings \> Chunk settings \> “** …” **\> Submit ad-hoc task.** From a branch that already exists, you can ask Chunk to accomplish any task you’d like (ie. “remove the outdated call-to-action from my web app’s home page”). It will push its changes to the branch that you select. For these tasks, Chunk runs in the environment that you define in your cci-agent-setup.yml file (read more about Chunk’s environment [here](https://discuss.circleci.com/t/product-launch-chunk-tasks-fixing-flaky-tests/53975/3#p-107197-unable-to-run-verification-tests-10)). ## Join the Beta **[Join the waitlist](https://www2.circleci.com/chunk-waitlist)** to get access > Have feedback or feature requests? Submit them on our **[Ideas board](https://circleci.canny.io/cloud-feature-requests?selectedCategory=chunk-ai)** where you can also see existing feature requests and vote on them. * * * **No extra cost during beta.** Uses compute credits and your AI provider tokens. This will be a paid feature after beta. ---
**Author:** ![luisejrobles](https://sea2.discourse-cdn.com/circleci/user_avatar/discuss.circleci.com/luisejrobles/32/36575_2.png) [@luisejrobles](https://discuss.circleci.com/u/luisejrobles)\ **Post date:** [September 5, 2025, 5:08pm UTC](https://discuss.circleci.com/t/product-launch-chunk-tasks-fixing-flaky-tests/53975/2 "2025-09-05T17:08:19Z")
## Chunk Tasks - Latest updates (09/05/2025) We’ve been actively improving the agent based on your feedback. Here are the latest enhancements: - When Assigning an Agent task, you can now configure the **Maximum number of concurrent open PRs** for the project. ## **Better User Experience** - We’ve improved agent logs within Agent tasks to be more human-friendly and conversation-like. Logs now clearly distinguish between **Assistant** (instructions being executed by the agent) and **User** (output from the agent), making it easier to follow the agent’s workflow. - In the Agent task Logs tab, we’ve added an **Expand/Collapse All** toggle to streamline troubleshooting. This allows you to quickly expand all logs and use Cmd+F (or Ctrl+F) to search for specific words or commands. - We’ve enhanced PR bodies and run summaries with **human-readable descriptions** for better clarity. - Agent-generated branches now use `circleci/fix-flaky-test-` instead of the previous `fix-flaky-test-` format. Making them easier to identify and filter in your repository. ### **Behind the Scenes** - **Enhanced Execution Environment** : The agent now uses a machine executor for better performance and reliability when running tests. > Have feedback or feature requests? Submit them on our **[Ideas board](https://circleci.canny.io/cloud-feature-requests?selectedCategory=chunk-ai)** where you can also see existing feature requests and vote on them. ---
**Author:** ![luisejrobles](https://sea2.discourse-cdn.com/circleci/user_avatar/discuss.circleci.com/luisejrobles/32/36575_2.png) [@luisejrobles](https://discuss.circleci.com/u/luisejrobles)\ **Post date:** [September 30, 2025, 11:49pm UTC](https://discuss.circleci.com/t/product-launch-chunk-tasks-fixing-flaky-tests/53975/3 "2025-09-30T23:49:41Z")
# Chunk - FAQs > Have feedback or feature requests? Submit them on our **[Ideas board](https://circleci.canny.io/cloud-feature-requests?selectedCategory=chunk-ai)** where you can also see existing feature requests and vote on them. ## Does CircleCI use my data to train the models? Your source code is not stored nor used for training purposes by CircleCI ### **What data does Chunk access?** Chunk accesses historical build data and repository contents to identify and fix flaky tests. This is the same information that CircleCI already has access to through your existing CircleCI configuration. ### **Will my test results be shared with other customers?** Your usage data, including test results, will not be used for any other customer. Each customer’s data remains isolated and is only used to support their own Chunk tasks. * * * ## How long are agent logs stored? We store agent logs for 90 days. This is a fixed retention period that applies to all organizations, regardless of [your plan’s standard data retention policy](https://support.circleci.com/hc/en-us/articles/5645222646939-Cloud-Data-Retention-Policy-and-Settings). After 90 days, logs are automatically deleted to keep your workspace at optimal performance. * * * ## OpenAI organization verification required. Please verify your organization at… When encountering the message: **`OpenAI organization verification required. Please verify your organization at `[`https://platform.openai.com/settings/organization/general`](https://platform.openai.com/settings/organization/general)`and see our`[`community forum`](https://discuss.circleci.com/t/product-launch-agentic-capability-fixing-flaky-tests/53975/2)` for more debugging help`** inside an agent task, it indicates that your OpenAI organization verification is still pending. **To fix this** : In [OpenAI Platform](https://platform.openai.com/settings/organization/general) navigate to **`General > Organization settings`** and click ‘Verify Organization’ to follow the necessary steps to have your organization verified. ![verifyorganization-ezgif.com-video-to-gif-converter](https://global.discourse-cdn.com/circleci/original/3X/7/b/7bf515b8cbc4c42d0b81cebdc1d31f57edcb4641.gif) **Additional help:** [OpenAI Organization Verification Guide](https://help.openai.com/en/articles/10910291-api-organization-verification) * * * ## What if I can’t get my organization verified when using OpenAI? If organization verification isn’t possible, you can bypass this requirement by adding an environment variable: 1. Go to **`Organization Settings > Contexts > circleci-agents`** 2. Add new Environment Variable: ![gpt-5-nano-ezgif.com-video-to-gif-converter (1)](https://global.discourse-cdn.com/circleci/original/3X/4/d/4d5afbd1d6b7c57ca8219a3ccbd0a021a0f0d6b5.gif) * * * ## Invalid OpenAI model specified. Please check the model name and ensure it is available for your account. When encountering the message: **`Invalid OpenAI model specified. Please check the model name and ensure it is available for your account.`** you need to make sure your organization has gpt-5 model access. **To verify this:** In [OpenAI Platform](https://platform.openai.com/settings/organization/general) 1. Switch to the project you want to check (top-left dropdown). 2. Go to **Settings → Limits** in the left-hand menu. 3. You can also check in the **Playground** : ![limitsss-ezgif.com-video-to-gif-converter](https://global.discourse-cdn.com/circleci/original/3X/7/b/7b1764414e81528e595f1caa790fe8ab1d147dbf.gif) * * * ## Action required - agent execution error When encountering the message: **`Action required - agent execution error`** **`The agent ran into an error while executing this task. See our community forum for how to solve this error.`** Email us at [sebastian@circleci.com](mailto:sebastian@circleci.com) and we’ll help you figure it out. * * * ## Unable to run verification tests Chunk runs in a[Linux Machine VM](https://circleci.com/docs/guides/execution-managed/using-linuxvm/)with [basic software](https://discuss.circleci.com/t/ubuntu-20-04-22-04-24-04-q4-edge-release/52429) installed by default. To verify that a proposed fix resolves flakiness, it re-runs the affected test several times. To do this, the agent may install additional software needed to set up the test environment, using clues from your **`circleci/config.yml`** to determine how to run the tests. You can view these attempts in the CircleCI web app by opening the **Chunk Tasks → Select a task logs → Expand All** , then searching for “run the command for each attempt.” This will take you to the sections where the agent is trying to run the tests. **Improving verification success** Create an “agent environment” CircleCI YML file. This file lets you copy the environment-setup parts of your existing CircleCI config into a dedicated file for Chunk. Name the file `cci-agent-setup.yml` and ensure that it is present in your `.circleci` directory and on the default branch. **Chunk supports all standard CircleCI configuration options.** This includes executors, resource classes, caching, contexts, environment variables, service containers, orbs, and everything else you’d use in a normal CircleCI pipeline. If it works in your `.circleci/config.yml`, it works in `cci-agent-setup.yml`. For a complete reference of available configuration options, see the[CircleCI Configuration Reference](https://circleci.com/docs/configuration-reference/). **Example `cci-agent-setup.yml` files:** > **Basic Python setup** > > ```yaml > version: 2.1 > workflows: > cci-agent-setup: > jobs: > - cci-agent-setup > jobs: > cci-agent-setup: > docker: > - image: cimg/python:3.12 > - image: cimg/postgres:15.3 > steps: > - checkout > - run: > name: Install dependencies > command: | > pip install -r requirements.txt > > ``` > **With Caching and Contexts** > > ```yaml > version: 2.1 > workflows: > cci-agent-setup: > jobs: > - cci-agent-setup: > context: > - my-team-context # Includes any secrets/env vars from this context > jobs: > cci-agent-setup: > docker: > - image: cimg/node:18.0 > steps: > - checkout > - restore_cache: > keys: > - v1-dependencies-{{ checksum "package-lock.json" }} > - run: > name: Install dependencies > command: npm install > - save_cache: > paths: > - node_modules > key: v1-dependencies-{{ checksum "package-lock.json" }} > > ``` > **With multiple services** > > ```yaml > version: 2.1 > workflows: > cci-agent-setup: > jobs: > - cci-agent-setup > jobs: > cci-agent-setup: > docker: > - image: cimg/ruby:3.2 > - image: cimg/postgres:15.3 > environment: > POSTGRES_USER: circleci > POSTGRES_DB: test_db > - image: redis:7.0 > steps: > - checkout > - run: > name: Wait for DB > command: dockerize -wait tcp://localhost:5432 -timeout 1m > - run: > name: Install dependencies > command: bundle install > - run: > name: Setup database > command: bundle exec rake db:setup > > ``` > **With custom resource class and machine executor** > > ```yaml > version: 2.1 > workflows: > cci-agent-setup: > jobs: > - cci-agent-setup > jobs: > cci-agent-setup: > machine: > image: ubuntu-2204:2024.01.2 > resource_class: large > steps: > - checkout > - run: > name: Install dependencies > command: | > sudo apt-get update > sudo apt-get install -y build-essential > > ``` ### **Environment Variables & Contexts** **Project environment variables:** Chunk automatically has access to any[environment variables you’ve configured at the project level](https://circleci.com/docs/set-environment-variable/#set-an-environment-variable-in-a-project) in CircleCI. You don’t need to recreate or reference these, they’re already available. **Contexts:** If you’re using[CircleCI contexts](https://circleci.com/docs/contexts/) to manage secrets or environment variables, simply include the context in your cci-agent-setup job (as shown in the caching example above). Chunk will have access to all variables from that context, no need to manually recreate them. ### **Testing Your Environment Setup** To build & iterate on Chunk’s environment, navigate to **Organization Settings → Chunk Tasks → Identify desired Agent Task → Select […] → Select [Chunk Environment]**. This page lets you run the contents of your `cci-agent-setup.yml` file on a specific branch and immediately see the results from those ad-hoc tasks. Use the **Custom** button to submit a task to Chunk and see the results. Merge the `cci-agent-setup.yml` file to your default branch when the results on the environment setup page are satisfactory. ### **Additional Guidance for Chunk** To improve Chunk’s ability to run tests & produce fixes that are aligned with stylistic/architectural preferences, many users also include a markdown file (`claude.md` or `agents.md`) in the root of their repo with instructions for running tests. Chunk should pick this up automatically. * * * ## Changing Chunk’s model provider Currently, Chunk can only have one model provider installed at a time. To change your model provider: 1. Navigate to **Organization settings** → **Chunk tasks** from the left navigation 2. Click **Edit in Contexts** 3. Select **circleci-agents** , scroll down to **Environment variables** and delete your current model provider API key 4. Click on **Add environment variable** and input your new API key information: * * * ## Task Summary or Pull Request Body Too Long or Poorly Formatted If you’re noticing that your Chunk Task responses appear incomplete or poorly formatted, this may indicate that your API key needs to be configured for a more capable model. **Identifying the Issue** When viewing a Chunk Task through the CircleCI UI → Chunk Tasks, you might observe these indicators of suboptimal model performance: - **Inconsistent formatting** : The task body lacks properly bolded section headers for **Run Summary** , **Root Cause** , **Proposed Fix** , and **Verification** - **Interactive prompts** : Chunk Task body ends with open-ended questions like: > “Would you like me to implement the robust wait pattern in the test now, and add a small helper for future tests? If yes, I’ll apply the changes and run the targeted tests.” To guide you, here are some examples of how a Chunk Task body should look: ![Good formatted Chunk Task summary example (1)](https://global.discourse-cdn.com/circleci/original/3X/2/4/241fa20e3f6af27c1c3eedecdb0ba6fc762e366a.png) ![Good formatted Chunk Task summary example (2)](https://global.discourse-cdn.com/circleci/original/3X/e/4/e4045aeb4c4c3ee33fe70c456c4359db9db09d35.png) **Root Cause** These formatting and content issues typically occur when Chunk is using a less powerful language model to analyze and propose fixes for flaky tests. **Resolution** To fix this, you’ll need to ensure your organization has access to gpt-5 and it’s properly verified. For details on verification requirements, see “OpenAI organization verification required. Please verify your organization at…” in our FAQs. **Important:** If your team previously overrode the model used by Chunk, you’ll need to remove that configuration to prevent using a lower-performance model: 1. Navigate to **[CircleCI web app](https://app.circleci.com/home) \>** **Organization Settings \> Contexts \> circleci-agents** 2. Remove the **`CCI_AGENT_OPENAI_MODEL`** environment variable from **Environment variables** section * * * ## Start Task button disabled We’ve noticed some users experiencing an issue where the **Start Task** button remains disabled in **Chunk Tasks \> Assign new task** , even after filling in all required inputs. **Resolution** 1. Look for a callout in the top navigation bar that says **“Your GitHub identity is not verified”** and click **Authorize.** [![image (1)](https://global.discourse-cdn.com/circleci/original/3X/a/8/a8c88eaafbd53bdda4bd66c134389231719e3918.png)](https://global.discourse-cdn.com/circleci/original/3X/a/8/a8c88eaafbd53bdda4bd66c134389231719e3918.png "image (1)") 1. Select **“Authorize CircleCI App”** when asked3 2. Try assigning a new Chunk Task again This should resolve the issue and allow you to proceed with assigning a Chunk task to a project. ---
**Author:** ![luisejrobles](https://sea2.discourse-cdn.com/circleci/user_avatar/discuss.circleci.com/luisejrobles/32/36575_2.png) [@luisejrobles](https://discuss.circleci.com/u/luisejrobles)\ **Post date:** [October 16, 2025, 8:24pm UTC](https://discuss.circleci.com/t/product-launch-chunk-tasks-fixing-flaky-tests/53975/6 "2025-10-16T20:24:03Z")
# Chunk Tasks - Latest Updates (10/16/2025) > Have feedback or feature requests? Submit them on our **[Ideas board](https://circleci.canny.io/cloud-feature-requests?selectedCategory=chunk-ai)** where you can also see existing feature requests and vote on them. ## Better User Experience - Chunk now uses failed job step context to improve fix accuracy. - We’ve fixed an issue where the Chunk Tasks nav bar item wasn’t highlighting when selected, improving navigation clarity. - We’ve resolved a bug where pasting an API key would automatically close the setup modal, ensuring a smoother setup experience. - **Chunk’s commits now include verified signatures**. All commits created by Chunk are now properly signed and authored by circleci-app[bot]. This resolves issues where unsigned commits would require special handling or higher privileges to merge. - You can now guide Chunk with a custom instruction file. Create a `fix-flaky-test.md` file in your `.circleci/` directory to provide specific guidance about how you want the agent to approach fixing flaky tests in your project. This gives you fine-grained control over the agent’s behavior and lets you encode your team’s testing best practices directly into the fix generation process. Example `.circleci/fix-flaky-test.md` file: ```markdown ## Command Restrictions - You MUST NOT use the `sleep()` command or `setTimeout()` for delays in any scripts - You MUST NOT use `eval()` as it poses security risks - Avoid using shell wildcards in destructive operations (e.g., `rm -rf *`) ## Code Style Preferences - Prefer functional components over class components in React - Use TypeScript `type` definitions instead of `interface` (this project enforces this via ESLint) - Favor explicit error handling over try-catch-all patterns - Use async/await syntax over Promise chains for readability ## Security Considerations - Always flag use of `dangerouslySetInnerHTML` in React components - Highlight any potential SQL injection vulnerabilities - Point out hardcoded credentials or API keys - Flag any use of `eval()` or `Function()` constructors ## Documentation Standards - Complex algorithms MUST include explanatory comments ``` ## Enhanced Functionality - Chunk is now using the latest model from Anthropic by default: Claude Sonnet 4.5 - The cci-agent-setup.yml configuration now works seamlessly with orbs and user-specified resource classes, giving you more flexibility in how you set up your agent environment. ## Behind the Scenes - We’ve added file protection safeguards to ensure the agent respects and excludes sensitive files from commits - We’ve restored full detail in execution logs when using OpenAI as model provider. A regression that caused Chunk execution logs to show significantly less output has been resolved. When in `CCI web app > Chunk Tasks > Chunk Task`, Logs now provide complete visibility into the agent’s actions, making troubleshooting much easier when reviewing task history. ---
**Author:** ![luisejrobles](https://sea2.discourse-cdn.com/circleci/user_avatar/discuss.circleci.com/luisejrobles/32/36575_2.png) [@luisejrobles](https://discuss.circleci.com/u/luisejrobles)\ **Post date:** [October 29, 2025, 8:25pm UTC](https://discuss.circleci.com/t/product-launch-chunk-tasks-fixing-flaky-tests/53975/7 "2025-10-29T20:25:36Z")
# Chunk Tasks - Latest Updates (10/29/2025) ## Better User Experience - We’ve improved visibility in Chunk Settings so you can now see all projects with scheduled tasks, whether you’re following them or not. Previously, the task configurations table only displayed projects you were actively following, which could make it harder to understand which projects had scheduled Chunk tasks configured across your organization. To delete tasks for both followed and unfollowed projects,you’ll need `Project Admin` or `Organization Admin` permissions. > For a complete reference of roles and permissions, see [Roles and permissions](https://circleci.com/docs/guides/permissions-authentication/roles-and-permissions-overview/). - **We’ve streamlined Chunk’s environment customization with one-click setup file creation.** When Chunk doesn’t find a cci-agent-setup.yml file in your repository, it will attempt to create one for you automatically. You can also manually generate one by navigating to **Organization Settings → Chunk Tasks** in the CircleCI web app, selecting the **[…]** menu for your desired task, then going to **Chunk Environment**. From there, simply choose your branch and select **[Create File in GitHub]** to generate the file directly in your repository ![Chunk Environment Setup automatically suggests cci-agent-setup.yml](https://global.discourse-cdn.com/circleci/original/3X/3/f/3fd038dc16189c711f9f8b94f6e97a606dc05b81.png) - **We’ve fixed a bug that removed the “More Details” section from Chunk’s pull request descriptions.** This section now includes a direct link to view the full task execution in the CircleCI web app, making it easier to review Chunk’s work, examine logs, and understand the complete context behind each fix. ![More details inside Chunk's PR's restored](https://global.discourse-cdn.com/circleci/original/3X/2/2/22bcfb6c23c83eca44510ab3fd5704bec1b7cd28.png) - **We’ve improved Chunk’s failure handling when both verification tests fail and pipeline monitoring encounters issues.** Chunk now avoids opening pull requests when it can’t verify the fix worked, preventing potentially broken PRs from appearing in your repository. - **Enhanced ad-hoc task capabilities**. Chunk now commits changes directly from ad-hoc prompts. When you submit an ad-hoc task through Organization Settings \> Chunk Settings, Chunk will push its changes to your selected branch, making it easier to iterate on custom tasks. ![Screenshot 2025-10-30 at 10.51.33 a.m. (1)](https://global.discourse-cdn.com/circleci/original/3X/1/c/1c38358bc4cd25713db8e13ed67e7bd83c0504f4.png) - **You’ll now see visual feedback in both Ad Hoc Tasks and the Chunk Environment results view as Chunk works through your custom instructions** , making it easier to understand how long tasks will take. ![Screenshare-2025-10-294_08_28PM-ezgif.com-video-to-gif-converter](https://global.discourse-cdn.com/circleci/original/3X/7/b/7b59243cb95571f4457c4d13001beec92aa0ad6a.gif) - **Custom instruction files now supported in ad-hoc tasks and Chunk Environment**. Chunk now automatically picks up guidance from `agents.md`, `claude.md`, and `fix-flaky-tests.md` files when running ad-hoc tasks and environment setup. Simply add these files to the root of your repository. - **We’ve added a search filter to the branch dropdown in ad-hoc tasks**. Instead of scrolling through all branches, you can now type to quickly find the branch you need making it much easier to work with repositories that have many branches. ![ScreenRecording2025-11-03at12.37.54p.m.-ezgif.com-video-to-gif-converter](https://global.discourse-cdn.com/circleci/original/3X/5/b/5b29905f48f9c6eaaa50dec836d0ba7f14585071.gif) # Chunk Tasks - Latest Updates (11/06/2025) > Have feedback or feature requests? Submit them on our **[Ideas board](https://circleci.canny.io/cloud-feature-requests?selectedCategory=chunk-ai)** where you can also see existing feature requests and vote on them. - **Chunk now validates changes by running your CI pipeline.** For Ad-Hoc Tasks, after Chunk pushes changes to a branch, it triggers and monitors your CI pipeline to verify the changes. If the pipeline fails, Chunk will attempt to fix the issues and push updated changes. > You can access Ad-Hoc Tasks by navigating to **Organization Settings → Chunk Tasks** , selecting the **‘⋮’ symbol** in the Task configurations table, then **Submit Ad Hoc Task**. > > ![Screenshot 2025-11-06 at 3.31.39 p.m. (1)](https://global.discourse-cdn.com/circleci/original/3X/c/e/ce002e6d5eb8871988db2dca90661f92f204fbe7.png) # Chunk Tasks - Latest Updates (11/18/2025) - We’ve improved Chunk’s logic to prevent re-opening PRs for tests that have already been successfully fixed. Once Chunk has addressed a flaky test and the fix has been merged, it will no longer attempt to fix that same test again, reducing unnecessary PRs and noise in your repository. - **Error message implemented for when running low on credits for Anthropic users.** Chunk now displays visual warnings throughout the UI when your Anthropic account credits are running low, helping you avoid unexpected task failures. You’ll see an “Insufficient Anthropic Credits” indicator on the Chunk Tasks dashboard, in Organization Settings, and on failed task detail pages. The warning includes clear steps to add credits at [console.anthropic.com](http://console.anthropic.com) and automatically clears once your next scheduled run, ad-hoc task, or environment setup completes successfully with sufficient credits. - **Chunk shows “Model restricted”** If you’re seeing a “Model restricted” chip or error message stating “Cannot diagnose fix due to OpenAI Zero Data Retention policy” in your Chunk Tasks, this means your OpenAI organization has Zero Data Retention enabled, which prevents Chunk from analyzing your test data and generating fixes. ![Captura de pantalla 2025-11-20 a la(s) 11.03.43 a.m. (1)](https://global.discourse-cdn.com/circleci/original/3X/1/b/1b4b4bf8a7dd85ae2ace60d72e5b2f056af082a4.png)