Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Thin dual-axis router for preset questions: which of the 12 TUI-shipped built-in preset templates (provider axis), and which cross-cutting preset lifecycle mechanic (operation axis) — saving, checking availability, activating/refreshing, endpoint/capability facts, or troubleshooting. Read a child only when it is relevant; this does not describe arbitrary saved presets.
.claude/skills/lingtai-ai-lingtai-preset-skill/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 42% | 0% |
This router covers exactly the names returned by BuiltinPresets() in tui/internal/preset/preset.go, plus the cross-cutting mechanics that apply across all of them. It is for TUI-owned template presets only.
Preset questions split along two independent axes:
endpoint, credential env-var, official links). One child per BuiltinPresets() name under reference/<provider>/SKILL.md.
/ troubleshoot a preset, regardless of provider?" One child per mechanic under reference/operations/<operation>/SKILL.md.
A concrete question usually composes both: read the operation child for the mechanic, then the provider child for the provider-specific fact (exact model, endpoint, credential). Provider pages own volatile model, pricing, endpoint, protocol, and plan facts and route to the relevant operation child instead of restating shared mechanics; operation pages own the shared lifecycle mechanics and never encode provider-specific facts.
yaml- name: preset-skill-minimax location: reference/minimax/SKILL.md - name: preset-skill-zhipu location: reference/zhipu/SKILL.md - name: preset-skill-mimo location: reference/mimo/SKILL.md - name: preset-skill-deepseek location: reference/deepseek/SKILL.md - name: preset-skill-gemini location: reference/gemini/SKILL.md - name: preset-skill-kimi location: reference/kimi/SKILL.md - name: preset-skill-grok location: reference/grok/SKILL.md - name: preset-skill-nvidia location: reference/nvidia/SKILL.md - name: preset-skill-openrouter location: reference/openrouter/SKILL.md - name: preset-skill-codex location: reference/codex/SKILL.md - name: preset-skill-codex-pool location: reference/codex-pool/SKILL.md - name: preset-skill-claude location: reference/claude/SKILL.md - name: preset-skill-custom location: reference/custom/SKILL.md
| Preset | Child manual | Stable routing hint | |---|---|---| | minimax | reference/minimax/SKILL.md | MiniMax, Anthropic-shaped endpoint | | zhipu | reference/zhipu/SKILL.md | Zhipu/Z.AI GLM Coding Plan | | mimo | reference/mimo/SKILL.md | Xiaomi MiMo, OpenAI-compatible | | deepseek | reference/deepseek/SKILL.md | DeepSeek, OpenAI-compatible text route | | gemini | reference/gemini/SKILL.md | Google Gemini native multimodal route | | kimi | reference/kimi/SKILL.md | Kimi Code, OpenAI-compatible | | grok | reference/grok/SKILL.md | Grok (xAI) via OpenCode Go, OpenAI-compatible | | nvidia | reference/nvidia/SKILL.md | NVIDIA NIM/API Catalog gateway | | openrouter | reference/openrouter/SKILL.md | OpenRouter gateway | | codex | reference/codex/SKILL.md | ChatGPT OAuth Codex route | | codex-pool | reference/codex-pool/SKILL.md | pooled ChatGPT OAuth Codex route | | claude | reference/claude/SKILL.md | local Claude Code login | | custom | reference/custom/SKILL.md | user-supplied compatible endpoint |
yaml- name: preset-skill-op-saved-presets location: reference/operations/saved-presets/SKILL.md - name: preset-skill-op-endpoint-capabilities location: reference/operations/endpoint-capabilities/SKILL.md - name: preset-skill-op-availability-save-gate location: reference/operations/availability-save-gate/SKILL.md - name: preset-skill-op-activation-session-refresh location: reference/operations/activation-session-refresh/SKILL.md - name: preset-skill-op-troubleshooting-migration location: reference/operations/troubleshooting-migration/SKILL.md
| Question shape | Child manual | |---|---| | How do saved presets differ from templates? Load/Save/Delete/Bootstrap order and atomicity? | reference/operations/saved-presets/SKILL.md | | What base URL / API compatibility / provider / model / capability declarations does a preset carry, distinct from credentials? Codex OAuth quota inspection. | reference/operations/endpoint-capabilities/SKILL.md | | Does Save make a live provider call, and where does real availability diagnosis live? | reference/operations/availability-save-gate/SKILL.md | | How does a saved preset become the running default, and what does /refresh actually switch? | reference/operations/activation-session-refresh/SKILL.md | | Something looks broken/stale — bounded triage before routing to a deeper runtime/update/migration skill | reference/operations/troubleshooting-migration/SKILL.md |
The manuals describe direct provider-native or current-preset OpenAI-compatible vision conservatively. A missing or failed direct route is not a reason to switch providers, guess credentials, or auto-load/invoke an MCP. Reviewed official evidence positively identifies optional plan-level vision MCPs for Zhipu/Z.AI and MiniMax; their child manuals link them as explicit, manual-only methods. Unknowns remain unknown.
Saved presets are different: a clone under ~/.lingtai-tui/presets/saved/ may change provider, model, endpoint, credentials, or capabilities. Inspect its actual manifest instead of treating this catalog as coverage for it.
When BuiltinPresets() gains a new template name, the same change must add a reference/<name>/SKILL.md child and both provider-catalog entries. When a new cross-cutting mechanic is added, add a reference/operations/<name>/SKILL.md child and both operation-catalog entries instead of folding it into a provider page. The focused router test keeps the source list, embedded children, parent metadata, and extracted paths in bijection for both axes.
Other measured skills in the registry, with their headline benchmark lift.