Agent Tty

Terminal Automation with agent-tty

If agent-tty is not already available in the environment, fetch the current README.md from coder/agent-tty on GitHub and follow its installation instructions before continuing. Otherwise use agent-tty directly. Examples use jq for JSON parsing; any JSON-processing tool works. Prefer isolated homes, JSON envelopes, and renderer-backed artifacts so terminal workflows stay reviewable and reproducible.

Core Workflow

Every terminal or TUI automation task should follow this pattern:

1. Create an isolated home with --home.
2. Check prerequisites with doctor --json before screenshot or recording work.
3. Create a session with create --json.
4. Run setup commands with run instead of simulating long shell typing.
5. Wait on observable terminal state with wait instead of blind sleeps.
6. Inspect the current screen with snapshot.
7. Capture proof artifacts with screenshot or record export.
8. Destroy the session when finished.

AGENT_HOME="$(mktemp -d)"
agent-tty --home "$AGENT_HOME" doctor --json
SESSION_ID=$(agent-tty --home "$AGENT_HOME" create --json -- /bin/bash | jq -r '.result.sessionId')
agent-tty --home "$AGENT_HOME" run "$SESSION_ID" 'printf "ready\n"'
agent-tty --home "$AGENT_HOME" wait "$SESSION_ID" --text 'ready' --json
agent-tty --home "$AGENT_HOME" snapshot "$SESSION_ID" --format text --json
agent-tty --home "$AGENT_HOME" screenshot "$SESSION_ID" --json
agent-tty --home "$AGENT_HOME" record export "$SESSION_ID" --format webm --json
agent-tty --home "$AGENT_HOME" destroy "$SESSION_ID" --json

Essential Commands

# Environment and lifecycle
agent-tty --home <path> doctor --json
agent-tty --home <path> create --json -- /bin/bash
agent-tty --home <path> inspect <session-id> --json
agent-tty --home <path> destroy <session-id> --json

# In-session control
agent-tty --home <path> run <session-id> 'command here' --json
agent-tty --home <path> type <session-id> 'literal text' --json
agent-tty --home <path> paste <session-id> 'multiline payload' --json
agent-tty --home <path> send-keys <session-id> Enter Ctrl+C --json
agent-tty --home <path> batch <session-id> '[{"run":"htop","noWait":true},{"wait":{"screenStableMs":1000}}]' --json

# Observation and proof
agent-tty --home <path> wait <session-id> --text 'ready' --json
agent-tty --home <path> wait <session-id> --regex 'READY>$' --scope cursor-line --json
agent-tty --home <path> wait <session-id> --screen-stable-ms 1000 --json
agent-tty --home <path> snapshot <session-id> --format text --json
agent-tty --home <path> screenshot <session-id> --json
agent-tty --home <path> record export <session-id> --format webm --json

Common Patterns

Bootstrap a shell session

AGENT_HOME="$(mktemp -d)"
SESSION_ID=$(agent-tty --home "$AGENT_HOME" create --json -- /bin/bash | jq -r '.result.sessionId')
agent-tty --home "$AGENT_HOME" run "$SESSION_ID" 'pwd && ls -la' --json
agent-tty --home "$AGENT_HOME" snapshot "$SESSION_ID" --format text --json

snapshot and a matched wait carry an optional screenHash (a hash of the visible screen text). Compare it across calls to tell whether the visible screen actually changed instead of diffing full text; equal hashes mean identical visible content even when the event sequence advanced.

Drive an interactive CLI or TUI

Use batch to run an ordered sequence of input-and-wait steps in one call instead of separate run/wait/send-keys invocations. Each wait step is anchored to a Wait Baseline — it only observes screen state produced after the preceding input step, so the sequence cannot race ahead and match a stale screen. A batch stops at the first failed step by default (--keep-going attempts every step).

AGENT_HOME="$(mktemp -d)"
SESSION_ID=$(agent-tty --home "$AGENT_HOME" create --json -- /bin/bash | jq -r '.result.sessionId')
agent-tty --home "$AGENT_HOME" batch "$SESSION_ID" '[
  { "run": "<interactive-command>", "noWait": true },
  { "wait": { "screenStableMs": 1000 } },
  { "sendKeys": ["Down", "Down", "Enter"] },
  { "wait": { "text": "<expected-label>" } }
]' --json
agent-tty --home "$AGENT_HOME" screenshot "$SESSION_ID" --json

A wait can still match the echo of a just-typed command. Add "scope": "cursor-line" to a wait step (or --scope cursor-line on a standalone wait) to restrict text/regex matching to the row the cursor is on — once Enter is pressed the cursor moves past the echoed line, so the echo cannot match. This is ideal for prompts, which render at the cursor; rendered lines are right-trimmed of trailing spaces, so anchor a prompt displayed as READY> with READY>$. A standalone cursor-line wait issued right after an input command can still observe the pre-input screen (for example a repeated prompt in an echo-disabled application), so pass the input command's returned seq as --after-seq; batch wait steps are anchored automatically. For output that scrolls past the cursor, use a distinctive output token or screenStableMs.

Export reviewer-facing artifacts

AGENT_HOME="$(mktemp -d)"
SESSION_ID=$(agent-tty --home "$AGENT_HOME" create --json -- /bin/bash | jq -r '.result.sessionId')
agent-tty --home "$AGENT_HOME" run "$SESSION_ID" 'printf "artifact proof\n"' --json
agent-tty --home "$AGENT_HOME" wait "$SESSION_ID" --text 'artifact proof' --json
agent-tty --home "$AGENT_HOME" screenshot "$SESSION_ID" --json
agent-tty --home "$AGENT_HOME" record export "$SESSION_ID" --format asciicast --json
agent-tty --home "$AGENT_HOME" record export "$SESSION_ID" --format webm --json

Anti-Patterns

• Do not reach for tmux, screen, or ad hoc PTY wrappers first when agent-tty can provide an isolated, inspectable session.
• Do not rely on blind sleep calls when wait --text, wait --idle-ms, or wait --screen-stable-ms can observe terminal readiness directly.
• Do not bypass --json when another tool or agent needs machine-readable results.
• Do not use external screenshot tools as the primary proof path when agent-tty screenshot and agent-tty record export can produce renderer-backed artifacts tied to the session timeline.
• Do not leave sessions running after the task ends; destroy them explicitly.
• Do not rewrite public examples into repo-local development invocations; the public workflow should stay agent-tty ....