Tui Smoke Test With Tmux

TUI Smoke Test With tmux

Use this skill to validate the real terminal behavior of the Agent Claim Network TUI. It supports both the default smoke test and task-specific tmux scripts for user-described interaction flows.

The project TUI entrypoint is:

cargo run --quiet --bin acn -- --config config.toml

The TUI uses ratatui + crossterm, so verify it in a real tty. Prefer tmux capture files over screenshots because they can be inspected with rg, sed, and diffs.

The bundled runners that inspect TOML config use Python's standard library and require Python 3.11+ (tomllib). Their shared preflight reports this requirement before starting a TUI.

Workflow

1. Review the TUI change and decide the expected screen states or key flow.
2. Ensure tmux is installed. If an environment or dependency is missing, fix that environment issue first.
3. Source export_env.sh before verification when it exists or the run needs API/model variables.
4. For broad startup verification, run the bundled smoke script from the repo root:

.agents/skills/tui-smoke-test-with-tmux/scripts/tui_tmux_smoke.sh

5. For a specific user story, write a focused script using scripts/tui_tmux_lib.sh; use an existing regression script as the structural example.
6. Inspect the generated *.txt captures and stderr.log; use exact text assertions where stable.

Focused Verification

When the user asks whether a concrete TUI behavior is fixed, do not rely only on the default /help smoke test. Convert the behavior into a tmux flow:

1. Name the flow after the behavior, for example queue-cancel-resend.
2. Define checkpoints: initial, queued, after_cancel, after_resend, etc.
3. At each checkpoint, capture the pane to a task-specific directory under target/.
4. Assert stable visible markers with rg; avoid brittle assertions on session ids, timing-dependent model output, or full-screen whitespace.
5. Always assert stderr.log is empty and the tmux session is cleaned up.

For one-off verification, create the script in a task-specific directory under target/. For a reusable regression, add it under this skill's scripts/ directory. Source scripts/tui_tmux_lib.sh, which provides:

• tui_start: build if needed, create the runner, and start the fixed-size tmux session.
• tui_capture <name>: save the current pane as <name>.txt.
• tui_send_keys ...: send keys such as "/help" Enter, Escape, PageUp.
• tui_assert_contains <capture> <regex> <message>: fail if a capture lacks an expected marker.
• tui_assert_not_contains <capture> <regex> <message>: fail if a capture shows a forbidden marker.
• tui_assert_stderr_empty: fail on stderr output.

Example custom flow:

#!/usr/bin/env bash
set -euo pipefail

REPO_ROOT="$(git rev-parse --show-toplevel)"
cd "$REPO_ROOT"
source "$REPO_ROOT/.agents/skills/tui-smoke-test-with-tmux/scripts/tui_tmux_lib.sh"

TUI_CASE="queue-cancel-resend"
TUI_OUT_DIR="target/tui-flows/$TUI_CASE"
tui_start

sleep 3
tui_capture "initial"
tui_assert_contains "initial" "ACN|open|initializing" "TUI did not start"

# Replace these keys with the flow required by the user story.
tui_send_keys "第一条" Enter
sleep 1
tui_send_keys "第二条" Enter
sleep 1
tui_send_keys Escape
sleep 1
tui_capture "after_cancel"

tui_assert_contains "after_cancel" "第二条|turn cancelled" "cancel did not restore expected state"
tui_assert_stderr_empty

If a flow depends on long LLM turns, prefer a controlled test hook or a focused unit test for the exact state transition, then use tmux to verify the terminal shell of the behavior. Do not fake successful external behavior in production code.

Default Checks

The bundled script builds --bin acn first, then starts a fixed-size tmux session and captures:

• initial.txt: startup screen.
• after_help.txt: screen after typing /help and pressing Enter.
• final.txt: final visible screen before cleanup, if the session still exists.
• stderr.log: stderr from the TUI process.

Treat the smoke test as passing only when:

• the captured screen shows the ACN status line or initialization state;
• /help shows the ACN commands header and the /skills command;
• stderr.log is empty;
• the tmux session is cleaned up.

Expected stable strings in the current TUI include Agent Claim Network, Whisper your wish here..., type / for commands · Enter sends, ACN commands, /skills, initializing, and open. Agent ids, session ids, elapsed time, model output, and terminal whitespace are dynamic and must not be treated as stable assertions.

Useful Commands

Run the default smoke test:

.agents/skills/tui-smoke-test-with-tmux/scripts/tui_tmux_smoke.sh

Run focused resize regressions:

.agents/skills/tui-smoke-test-with-tmux/scripts/tui_tmux_welcome_resize_regression.sh
.agents/skills/tui-smoke-test-with-tmux/scripts/tui_tmux_live_region_regression.sh

Use a different agent or config:

.agents/skills/tui-smoke-test-with-tmux/scripts/tui_tmux_smoke.sh \
  --command "cargo run --quiet --bin acn -- --config config.toml --upstream agent_hub"

自定义命令默认不管理 supervisor，避免终止共享 config 下原本存在或后来复用该 supervisor 的其他 TUI。只有 config 与 acn_home 都是本次测试独占的临时资源时， 才传 --supervisor-config 启用回收；其值必须与自定义命令传给 --config 的参数 文本一致。直接使用已构建二进制时还可传 --supervisor-binary /absolute/path/to/acn，加强进程候选过滤。

Skip the pre-build only when an explicit binary is already known to be current:

TUI_ACN_BINARY=/absolute/path/to/acn \
  .agents/skills/tui-smoke-test-with-tmux/scripts/tui_tmux_smoke.sh --skip-build

Keep ANSI escape codes for color/style review:

tmux capture-pane -t acn_tui_smoke -e -p > target/tui-smoke/screen.ansi.txt

Send keys manually to an active session:

tmux send-keys -t acn_tui_smoke "/help" Enter
tmux send-keys -t acn_tui_smoke PageUp PageDown
tmux send-keys -t acn_tui_smoke Escape
tmux send-keys -t acn_tui_smoke "/exit" Enter

Always clean up stale sessions before or after a failed run:

tmux kill-session -t acn_tui_smoke 2>/dev/null || true

When Debugging Failures

• Empty or garbled capture: confirm the command is running in tmux, the fixed size is large enough, and the TUI entered alternate screen.
• Missing prompt/footer: inspect src/session_tui/chat_widget.rs, src/session_tui/bottom_pane/mod.rs, and the current rendering tests.
• Keys not taking effect: inspect handle_key_event, classify_input, and whether the state accepts text.
• Non-empty stderr: fix the underlying panic, terminal error, config issue, or dependency problem instead of bypassing the smoke test.
• Startup hangs: capture the initial screen and stderr, then decide whether the app is waiting on external services, LLM env vars, or session initialization.