# Launch Agents and Commands in New Terminals

Interactive tools and long-running commands can block the current coding session. This skill opens them in separate terminals and optionally captures their output.

## Install

```bash
npx skillstore add consiliency/spawn
```

## Metadata

- Status: approved
- Slug: consiliency-spawn
- Skillstore revision: r1
- Version status: missing
- Tree hash: a92bcb1c88fe565c8883adb9bf6c07b1bc13ca512914608e5abd34f6470511fd
- Author: Consiliency
- GitHub username: Consiliency
- License: MIT
- Repository: https://github.com/Consiliency/treesitter-chunker/tree/main/.ai-dev-kit/skills/spawn
- Ref: 635f69fb8d2f4e6330ba47a4e5a0fb239c04d110
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: blocked
- Manual install advisory: allowed\_with\_warning
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: env\_access, external\_commands, filesystem
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/consiliency-spawn
- Manifest: https://skillstore.pages.dev/api/skills/consiliency-spawn/manifest

## Capabilities

- Spawns supported AI CLIs in new terminal windows using provider-specific cookbooks.
- Runs generic CLI commands in separate terminal sessions.
- Supports macOS Terminal, Windows Terminal, WSL, tmux, and a background-process fallback.
- Captures clean agent output or full debug output in temporary files.
- Detects selected provider authentication failures and opens login commands.
- Reads completed output with a configurable timeout.

## Use Cases

- Delegate a coding task: Launch a selected coding agent in another terminal while keeping the primary session available.
- Start a long-running command: Run builds, media processing, or diagnostics in a separate interactive terminal.
- Test provider workflows: Open provider CLIs, complete authentication, and capture responses for evaluation.

## Prompt Templates

### Run one command

```
Open a new terminal in {working_directory} and run {command}. First verify the command syntax with its help option.
```

### Delegate to one agent

```
Spawn {provider} in a new terminal for this task: {task}. Follow its cookbook and keep approval prompts enabled.
```

### Capture an agent response

```
Spawn {provider} for {task}, capture clean output, wait up to {timeout} seconds, then summarize the result and report the output location.
```

### Compare two agent reviews

```
Launch Claude Code and Codex separately to review {scope}. Use sandboxed modes, capture both responses, and compare agreements, differences, and unresolved risks.
```

## Limitations

- Executes constructed shell commands and does not validate or restrict their contents.
- Requires installed terminal applications, tmux, and requested CLI tools.
- Windows capture behavior is limited, and platform fallbacks differ.
- Temporary output and script files are not automatically removed or permission-hardened.

## Best Practices

- Confirm the exact command, working directory, and provider before launching.
- Use sandboxed modes and approval prompts unless the user explicitly authorizes broader access.
- Remove temporary output files promptly and avoid storing secrets in captured logs.

## Anti Patterns

- Do not pass untrusted text directly into shell commands.
- Do not disable sandbox and approval controls by default.
- Do not forward conversation history until sensitive values are removed and the user confirms.

## Security Audit

- Audited at: 2026-07-19T04:12:07.468\+00:00
- Summary: Most static alerts in Markdown, comments, UUID generation, and constant /proc reads are false positives. Confirmed risks include unsanitized shell command execution and shared temporary files that can expose agent output or generated scripts. Cookbook defaults also disable approval and sandbox controls, which can grant unrestricted system access.

## Stats

- Views: 222
- Downloads: 10
- Favorites: 0
- Popularity score: 0