> ## Documentation Index
> Fetch the complete documentation index at: https://docs.myrmagent.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Batch Directory Runs

> Run one prompt across many directories at once — with artifact verification, failure retry, approval gates, and completion notifications.

# Batch Directory Runs

Run a single prompt across **multiple working directories in parallel**. Perfect for auditing many repositories for security issues, applying the same migration across projects, or collecting a consistent report from every codebase you manage.

## Creating a Batch Run

Open **Batch Directories** in the sidebar and click **New Run**. Configure:

| Setting               | What it does                                                        |
| --------------------- | ------------------------------------------------------------------- |
| **Name**              | Identifier for the batch project                                    |
| **Prompt**            | One instruction applied to every directory                          |
| **Directories**       | Workspaces to process (one Agent task per directory)                |
| **Concurrency**       | How many directories are processed at once                          |
| **Agent & Model**     | Optional agent binding / per-run model override                     |
| **Max runtime**       | Per-directory timeout guard (seconds)                               |
| **Require approval**  | Gate every result behind a human approve/reject                     |
| **Artifact patterns** | Glob patterns the agent must produce (e.g. `**/security-report.md`) |
| **Notifications**     | In-app + IM channel alerts on completion                            |

Dangerous paths (home directory, `/etc`, filesystem root, `~`, `..` traversal) are rejected at creation, and duplicate directories are automatically deduplicated.

## Monitoring Progress

The detail page shows live status, per-directory chips, and aggregated counts. Directories whose task **failed** or **produced no required artifacts** are highlighted so you can spot problems at a glance. Total wall-clock duration is shown once the run finishes.

## Handling Failures

Instead of re-running everything, recover precisely:

* **Retry failed directories** — resend tasks only for directories that failed or are missing artifacts
* **Retry one directory** — re-run a single task
* **Rerun all** — requeue every directory (terminal projects only, to avoid duplicates)

Retries are tracked per directory generation, so a successful retry brings the whole project back to **completed** — even if earlier attempts failed.

## Pausing & Approval

* **Pause** halts running executions and freezes the queue; **resume** unfreezes exactly where you stopped.
* With **Require approval** enabled, every verified task lands in *in review* and waits for your approve/reject. A one-click **Approve all** finalizes the whole batch.

## Notifications

When the batch reaches a terminal state you get an in-app notification **and** an IM channel message (if the bound agent has notify targets) summarizing results — totals, duration, and the failed directory list. Messages include a deep link straight to the project detail page, so you can jump from chat to the full run.

## Typical Use Cases

* Audit N repositories for security or license issues with one prompt
* Apply the same refactor/migration across many projects
* Generate a consistent status report for every directory you own