Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Anything you need to know about LingTai TUI. Read this first for the stable mental model, major-feature map, interface and slash-command help, or the correct source/manual route for a TUI question. It stays thin: exact command prose lives in the localized help assets, precise Go structure lives in `lingtai-tui-anatomy` plus the repository ANATOMY graph, and independent domains such as presets, Portal, tutorials, updates, and addons keep their own top-level skills.
.claude/skills/lingtai-ai-lingtai-tui-help/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 47% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 71% | 0% |
lingtai-tui and lingtai-portal are control and presentation surfaces. The running agent is a separate LingTai kernel process with its own heartbeat, listeners, lifecycle, and durable working directory.
running agent. Reopening the TUI observes the current state and does not silently revive an intentionally stopped agent.
/sleep, /suspend, /cpr, or/refresh according to the intended transition; use lingtai-tui list --detailed or /projects to inspect current state.
launchd is not the default remedy for ordinary persistence after TUIexit. The agent already runs independently; adding another supervisor can create duplicate-process races. Use a separate scheduler/service contract only for a separately authorized advanced workflow.
lingtai-tui-anatomy, start at the repositoryroot ANATOMY.md, and descend to tui/ANATOMY.md or portal/ANATOMY.md. Kernel-side process, heartbeat, listener, and lifecycle internals belong to lingtai-kernel-anatomy; cross-repo references stay narrative.
Sanitized regression question: “关闭 TUI/终端后 agent 会不会停止?” No. UI exit is not an agent lifecycle command; the agent continues until an explicit lifecycle action or an external process failure stops it. Do not install a launchd job merely to keep an ordinary agent alive after closing the TUI.
This YAML is the machine-readable map. Each route names the first owner to load; next is optional deeper evidence.
yaml- feature: runtime-lifecycle-source route: lingtai-tui-anatomy next: lingtai-kernel-anatomy - feature: slash-commands-keyboard-interaction route: assets/slash-commands.en.md localized_routes: [assets/slash-commands.zh.md, assets/slash-commands.wen.md] - feature: presets-providers-capabilities route: lingtai-preset-skill - feature: portal-visualization-replay route: lingtai-portal-guide - feature: tutorial-orientation route: tutorial-guide - feature: install-update-build route: lingtai-update - feature: contributor-setup-source-troubleshooting route: lingtai-dev-guide - feature: runtime-health-doctor route: lingtai-doctor - feature: projects-and-running-agent-inventory route: assets/slash-commands.en.md localized_routes: [assets/slash-commands.zh.md, assets/slash-commands.wen.md] next: lingtai-tui-anatomy - feature: addons-and-external-channels route: mcp-manual
| Need | Start here | What remains authoritative | |---|---|---| | Does closing TUI stop an agent? How do attach, list, quit, or lifecycle controls work? | lingtai-tui-anatomy | Repo-root ANATOMY.md → tui/ANATOMY.md / portal/ANATOMY.md; kernel internals → lingtai-kernel-anatomy | | What does a slash command do? Which keys drive the interface? | assets/slash-commands.en.md or its ZH/Wen sibling | DefaultCommands() and the localized help assets | | How do presets, providers, model parameters, and capabilities work? | lingtai-preset-skill | That skill's provider and operation routes | | How do Portal topology, APIs, recording, and replay work? | lingtai-portal-guide | Portal guide, then Portal Anatomy/source for implementation | | I am new to LingTai or want the guided course | tutorial-guide | Its lesson sequence | | How do I install, update, build, or diagnose install method? | lingtai-update | Its focused update/install references | | How do I contribute, set up a dev checkout, or troubleshoot source? | lingtai-dev-guide | Its focused contributor/debug references | | Is an agent offline, unreachable, or inconsistent across process/heartbeat/log surfaces? | lingtai-doctor | Its layered read-only health diagnostics | | How do /projects and running-agent inventory work? | Localized slash-command asset | lingtai-tui list --detailed, /projects, then lingtai-tui-anatomy for source | | How do I configure addons or external messaging channels? | mcp-manual | The curated addon documentation it routes to |
The complete slash-command reference remains in three language assets so the in-app /help view can render the current UI language directly:
assets/slash-commands.en.md — English (canonical wording).assets/slash-commands.zh.md — 简体中文.assets/slash-commands.wen.md — 文言.Keep these assets in sync with DefaultCommands() in tui/internal/tui/palette.go. When a command is added, changed, or removed, update all three assets together.
/help resolves the languageThe top-level skill is the discoverable umbrella; the in-app /help command is a focused shortcut to the slash-command guide. It calls i18n.Lang() ("en", "zh", or "wen") and opens the matching concrete asset from the list above in the Markdown viewer. Unknown locales fall back to assets/slash-commands.en.md.
Keep general TUI orientation and feature routing here. Do not copy precise Go architecture from Anatomy, provider contracts from lingtai-preset-skill, Portal detail from lingtai-portal-guide, or tutorial/update/addon procedures from their owners. Route, then load only the source that owns the question.
Other measured skills in the registry, with their headline benchmark lift.