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

# CLI: Build Tasks

> Connect repositories and evals, and generate tasks from merged PRs or described vibe evals.

`R` is `--repo OWNER/NAME`, or another [scope](https://docs.selfbench.dev/cli/overview.md#conventions).

| Command | Does |
| - | - |
| `repos` | Connected repositories and evals |
| `repo connect\|show\|disconnect R` | Connect, read, or disconnect a repository |
| `repo set R --agent-minutes N` | How long an agent may work on each task |
| `eval create NAME` | A workspace eval; then use `--repo evals/NAME` |
| `prs R [--page N]` | Merged pull requests |
| `generate R --easy N --medium N --hard N [--focus TEXT]` | Start a batch that picks PRs and authors tasks |
| `task add-pr R --pr NUMBER\|URL` | A task from one merged PR |
| `scenario create R --title T (--scenario TEXT \| --scenario-file F) [--commit REF] [--attach F ...] [--difficulty D]` | A task from a described vibe eval |
| `batches R` · `batch show\|cancel RUN_ID R` | Follow or stop batches |

## Generation Options

`generate`, `task add-pr`, `scenario create`, and `task revise` take `--model`, `--verifier-model`, `--reasoning`, `--model-credential`, `--sandbox`, `--sandbox-credential`, `--harbor`, and `--harbor-credential`, or `--generation JSON|@FILE`. Without them, they use managed models and sandbox.

See [How It Works](https://docs.selfbench.dev/concepts/how-it-works.md) for what makes a task valid.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.