Agent skill
dev-terminal
Terminal/PTY automation with persistent sessions. Use to run and interact with TUI applications, debug terminal apps, automate CLI workflows, or any terminal interaction. Trigger phrases include "run the TUI", "start the app", "debug the terminal", "interact with", "send keys", "what's on screen".
Install this agent skill to your Project
npx add-skill https://github.com/parkerhancock/dev-terminal/tree/main/skills/dev-terminal
SKILL.md
Dev Terminal Skill
Terminal automation that maintains PTY sessions across script executions. Run TUI applications, send keystrokes, capture screen output, and debug terminal apps - all with persistent state.
Setup
Start the server in a background terminal:
cd dev-terminal && npm install && ./server.sh &
Wait for the Ready message before running scripts.
Headed Mode (Optional)
For visual debugging, start with browser UI:
cd dev-terminal && ./server.sh --headed &
This opens a browser window showing all terminals in real-time. Useful for:
- Watching AI actions as they happen
- Manual intervention if needed
- Debugging TUI interactions
Writing Scripts
Run scripts inline using heredocs from the dev-terminal/ directory:
cd dev-terminal && npx tsx <<'EOF'
import { connect, sleep } from "./src/client.js";
const client = await connect();
// Create or get a named terminal
const term = await client.terminal("my-app", {
command: "python",
args: ["-m", "my_module"],
cols: 120,
rows: 40,
});
// Wait for app to start
await sleep(1000);
// Get screen snapshot
const snap = await term.snapshot();
console.log("=== SCREEN ===");
console.log(snap.text);
client.disconnect();
EOF
Shell Defaults
By default, terminals use:
- Shell: User's default shell (
$SHELLenv var, e.g., zsh on macOS, bash on Linux) - Mode: Login shell (
-lflag) - loads full profile (~/.zprofile,~/.bash_profile)
This means your shell aliases, PATH, and environment are available.
Override the shell:
// Use a specific shell
const term = await client.terminal("my-term", {
command: "bash",
args: ["-l"], // keep as login shell
});
// Non-login shell (only loads ~/.bashrc, not ~/.bash_profile)
const term = await client.terminal("my-term", {
command: "bash",
args: [], // no -l flag
});
// Run a command directly (not a shell)
const term = await client.terminal("my-term", {
command: "python",
args: ["-m", "my_app"],
});
SSH Remote Terminals
Connect to remote servers via SSH. The API is identical to local terminals.
import { connect } from "./src/client.js";
import * as fs from "fs";
import * as os from "os";
import * as path from "path";
const client = await connect();
// SSH with private key
const term = await client.terminal("remote-server", {
ssh: {
host: "192.168.1.100",
username: "deploy",
privateKey: fs.readFileSync(path.join(os.homedir(), ".ssh/id_rsa"), "utf8"),
},
});
// SSH with password
const term = await client.terminal("remote-server", {
ssh: {
host: "example.com",
username: "admin",
password: "secret",
},
});
// SSH with agent (uses SSH_AUTH_SOCK)
const term = await client.terminal("remote-server", {
ssh: {
host: "example.com",
username: "admin",
agent: process.env.SSH_AUTH_SOCK,
},
});
// SSH with custom port and encrypted key
const term = await client.terminal("remote-server", {
ssh: {
host: "example.com",
port: 2222,
username: "admin",
privateKey: fs.readFileSync("/path/to/key", "utf8"),
passphrase: "key-passphrase",
},
});
SSH Options:
| Option | Type | Description |
|---|---|---|
host |
string | Remote hostname or IP (required) |
port |
number | SSH port (default: 22) |
username |
string | SSH username (required) |
password |
string | Password authentication |
privateKey |
string | Private key content (not path) |
passphrase |
string | Passphrase for encrypted keys |
agent |
string | Path to SSH agent socket |
Notes:
- SSH terminals don't have a
pid(it'sundefined) - All Terminal methods work the same (write, key, snapshot, etc.)
- The remote shell is determined by the server, not local settings
Key Principles
- Small scripts: Each script does ONE thing (start app, check screen, send key)
- Observe output: Always check
snapshot()to see current state - Descriptive names: Use
"claude-monitor","vim-edit", not"term1" - Terminals persist:
disconnect()leaves terminals running for next script
Workflow Loop
- Write a script to perform one action
- Run it and observe the screen output
- Evaluate - what's displayed? Did it work?
- Decide - send more input or task complete?
- Repeat until done
Client API
const client = await connect();
// Get or create named terminal (uses default shell as login shell)
const term = await client.terminal("name");
// With options
const term = await client.terminal("name", {
command: "python", // override shell/command
args: ["-m", "my_app"], // override args (clears default -l flag)
cols: 120,
rows: 40,
cwd: "/path/to/dir",
env: { MY_VAR: "value" },
});
// List all terminal names
const names = await client.list();
// Close/kill a terminal
await client.close("name");
// Disconnect (terminals persist)
client.disconnect();
Terminal Methods
// Send raw input
await term.write("hello");
// Send special keys
await term.key("enter");
await term.key("up");
await term.key("ctrl+c");
// Send a line (adds Enter)
await term.writeLine("ls -la");
// Get screen state
const snap = await term.snapshot();
console.log(snap.text); // Plain text (no ANSI codes)
console.log(snap.lines); // Array of lines
console.log(snap.alive); // Process still running?
// Get SVG rendering (for visual analysis)
const svgSnap = await term.snapshot({ format: "svg" });
console.log(svgSnap.svg); // SVG string
// Resize
await term.resize(80, 24);
// Clear buffer
await term.clear();
// Wait for text to appear
const found = await term.waitForText("Ready", { timeout: 5000 });
// Wait for process to exit
const exitCode = await term.waitForExit({ timeout: 10000 });
Special Keys
Use term.key() with these names:
| Category | Keys |
|---|---|
| Arrows | up, down, left, right |
| Control | enter, tab, escape, backspace, delete |
| Ctrl+X | ctrl+c, ctrl+d, ctrl+z, ctrl+l, ctrl+a, ctrl+e, ctrl+k, ctrl+u, ctrl+w, ctrl+r |
| Function | f1 - f12 |
| Navigation | home, end, pageup, pagedown, insert |
Example: Debug a TUI App
cd dev-terminal && npx tsx <<'EOF'
import { connect, sleep } from "./src/client.js";
const client = await connect();
// Start the TUI
const term = await client.terminal("claude-monitor", {
command: "../.venv/bin/python",
args: ["-m", "claude_monitor"],
cols: 120,
rows: 40,
cwd: "..",
});
// Wait for it to render
await sleep(2000);
// Capture screen
const snap = await term.snapshot();
console.log("=== SCREEN OUTPUT ===");
console.log(snap.text);
console.log("=== ALIVE:", snap.alive, "===");
client.disconnect();
EOF
Example: Interactive Session
# Script 1: Start app
cd dev-terminal && npx tsx <<'EOF'
import { connect, sleep } from "./src/client.js";
const client = await connect();
const term = await client.terminal("my-tui", {
command: "htop",
});
await sleep(1000);
const snap = await term.snapshot();
console.log(snap.text);
client.disconnect();
EOF
# Script 2: Send keys (terminal persists!)
cd dev-terminal && npx tsx <<'EOF'
import { connect, sleep } from "./src/client.js";
const client = await connect();
const term = await client.terminal("my-tui"); // Reconnect to existing
await term.key("down");
await term.key("down");
await sleep(500);
const snap = await term.snapshot();
console.log(snap.text);
client.disconnect();
EOF
# Script 3: Quit
cd dev-terminal && npx tsx <<'EOF'
import { connect } from "./src/client.js";
const client = await connect();
const term = await client.terminal("my-tui");
await term.write("q");
client.disconnect();
EOF
Error Recovery
If something goes wrong, check the terminal state:
cd dev-terminal && npx tsx <<'EOF'
import { connect } from "./src/client.js";
const client = await connect();
// List all terminals
const terminals = await client.list();
console.log("Active terminals:", terminals);
// Check specific terminal
if (terminals.includes("my-app")) {
const term = await client.terminal("my-app");
const snap = await term.snapshot();
console.log("Alive:", snap.alive);
console.log("Exit code:", snap.exitCode);
console.log("Last output:", snap.lines.slice(-20).join("\n"));
}
client.disconnect();
EOF
Tips
- TUI apps need time: Use
sleep()after starting to let them render - Check alive status: TUI might crash - check
snap.alive - Clear for fresh state: Use
term.clear()before important snapshots - Large output:
snap.linesgives the last ~120 lines (3x terminal height)
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
canvas
canvas
Render interactive UIs in a browser window using file operations. Supports two modes: React (App.jsx) for rapid prototyping with pre-built components, and Vanilla (index.html) for standards-based, portable artifacts. Use when users ask to: show forms, render charts/graphs, create dashboards, display data tables, build visual interfaces, or show any UI component. Trigger phrases: "show me", "render", "display", "create a form/chart/table/dashboard".
ip_research
IP data research tools for patents, trademarks, and applications. Use when: - Looking up patents by number (US, EP, WO, JP, etc.) - Searching patent databases by keyword, assignee, inventor, or classification - Getting patent family, citation, or legal status information - Checking USPTO application status, file wrapper, or PTAB proceedings - Downloading bulk USPTO data products - Finding patent assignments or ownership history - Fetching USPTO publication full-text data
edit-article
Edit and improve articles by restructuring sections, improving clarity, and tightening prose. Use when user wants to edit, revise, or improve an article draft.
obsidian-vault
Search, create, and manage notes in the Obsidian vault with wikilinks and index notes. Use when user wants to find, create, or organize notes in Obsidian.
handoff
Compact the current conversation into a handoff document for another agent to pick up.
Didn't find tool you were looking for?