# Agentrun CLI

> Execute shell commands with the local `agentrun` CLI. Use when an agent needs timeout control, bounded stdout and stderr capture, or structured command results instead of direct raw process execution.

- Skill: `sebastianboehler/agentrun-cli` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add sebastianboehler/agentrun-cli`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sebastianboehler/agentrun-cli/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: SebastianBoehler (https://skillmd.com/u/sebastianboehler)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/sebastianboehler/agentrun-cli

---


# AgentRun CLI

Use `agentrun` when direct command execution would be too noisy or risky because the process can hang, flood output, or return unstructured results.

## Invoke the Tool

Prefer the installed binary when it exists:

```bash
agentrun -timeout 5s -- sh -c 'echo ok'
```

If working inside this repository or the binary is not installed, run:

```bash
go run ./cmd/agentrun -- -timeout 5s -- sh -c 'echo ok'
```

## Choose Flags

- Use `-timeout` to cap command duration.
- Use `-max-output` to cap captured stdout and stderr bytes.
- Use `-dir` when the command must run in a specific directory.
- Use `-stdin` only if the child process must read caller input.
- Use `-format json` or `-format yaml` for programmatic consumers.
- Use `-format text` for quick inspection.

## Apply Good Defaults

- Keep short timeouts for probes and readiness checks.
- Keep small output caps for logs and diagnostic commands.
- Prefer `agentrun` over raw `sh -c` when the result needs to be parsed later.

## Examples

Run a health check:

```bash
go run ./cmd/agentrun -- -timeout 3s -format json -- sh -c 'uptime && df -h'
```

Capture a bounded log sample:

```bash
go run ./cmd/agentrun -- -timeout 2s -max-output 4096 -- journalctl -n 100
```