Agentic AI Notebook
Phase 26

Terminal Agent

~3 min read

Concept & How It Works

  • Key points are in the visual diagram above.

Why Does It Exist?

Much of engineering still happens in the terminal. Agents that can run commands accelerate exploration and automation, but require strict sandboxing to prevent destructive mistakes.

Real-World Analogy

An intern with a supervised shell — they can run commands you approve, in a disposable VM, with command blocklists for `rm -rf /`.
Loading diagram...

Visual Workflows

What is Terminal Agent?

Loading diagram...

Example

Scenario

User asks 'why is port 3000 in use?' The agent runs `lsof -i :3000`, identifies a stale node process, asks permission, then kills it and verifies the port is free.

Solution

In Coding Agents, apply Terminal Agent to this scenario: User asks 'why is port 3000 in use?' The agent runs `lsof -i :3000`, identifies a stale node process, asks permission, then kills it and verifies the port is free. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Practice Task

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Terminal Agent (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Code Walkthrough

Highlighted lines show where Terminal Agent happens in the code.

Terminal Agent
1async function runTerminalCommand(cmd, opts) {  # key line for Terminal Agent2  if (DENYLIST.some(d => cmd.includes(d))) throw new SecurityError("command blocked");3  if (REQUIRES_APPROVAL.some(p => cmd.startsWith(p)) && !opts.approved) {4    return { status: "pending_approval", command: cmd };  # return the result5  }6  const proc = spawn("bash", ["-c", cmd], { cwd: opts.cwd, timeout: 60_000 });7  const { stdout, stderr, exitCode } = await collectOutput(proc);8  audit.log({ event: "terminal.exec", cmd, exitCode });  # key line for Terminal Agent9  return { stdout: truncate(stdout, 50_000), stderr, exitCode };  # return the result10}

Commands to Remember

Commands to Remember

  • pip install openai # code generation and review
  • gh pr create # open a pull request from agent output

Common Mistakes

  • Treating Terminal Agent as a black box without evaluation
  • Ignoring cost and latency in production
  • Skipping error handling for terminal agent

Cheat Sheet

Quick recap — the most important points from this module.

Cheat Sheet

quick ref
  • Terminal Agent
  • Sandbox
  • Command Allowlist
  • Ephemeral Container