token-stack-setup

Automated setup, configuration, and provisioning engine across all 14 layers of the Token & Context Engine. Use for initial setup, component repair, and global CLI registration.

跳到安装

仓库: palpal2312/token-stack
最近来源活动: 2026年9月3日 15:32
检测到的 SKILL.md 语言: 英语
星标: 6
分支: 0

安装方式

默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

直接命令不会经过审查 Prompt；运行前请先检查来源。

npx skills add https://github.com/palpal2312/token-stack --skill token-stack-setup

命令会保持在同一行。复制前请横向滚动并检查完整内容。

想先保存到本地？可下载 SkillsMP 当前能够提供的文件。

检查来源文件

决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。

文件资源管理器

2 个文件

SKILL.md

来源说明 · 只读预览

返回搜索返回安装

token-stack

Unified router for the 14-Layer Master Token & Context Engine (Semantic Cache, Model Router, Skill Router, Code Topology, Data Lens, Ponytail, Caveman, RTK, Turn Folding, CoT Governor, Loop Breaker, Headroom, MemoraX, OpenViking).

2026-09-03

token-stack-benchmark

3-Step Interactive Benchmark Suite for the 14-Layer Token Stack (L-1 Semantic Cache, L0 Model Router, L0.5 Skill Router, L1 Topology, L1.5 Data Lens, L2 Ponytail, L3 Caveman, L4 RTK, L5 Turn Folding, L6 CoT Governor, L7 Loop Breaker, L8 Headroom, L9 MemoraX, L10 OpenViking). Features dataset presentation, layer toggles, N-run mean calculation, Dual Rubric evaluation, and Leave-One-Out Ablation Study across 12 public GitHub scenarios.

2026-09-02

token-stack-health

Live health checker and doctor diagnostic probes inspecting all 14 layers of the Token & Context Engine. Use for doctor, diagnostics, port readiness, and daemon checks.

2026-09-02

查看这个仓库的全部 Skills

Token Stack Setup Engine (14-Layer Automated Setup)

The setup engine inspects, provisions, configures, and verifies all required components across the 14-Layer Master Token & Context Engine:

📋 Configured Layers & Components

Layer -1: Semantic Cache: Provisions SQLite database at ~/.token-stack/cache.db with indexed schema.

Layer 0: Model Router: Provisions ~/.token-stack/router-config.json with multi-tier model cascading (cheap: kimi-k3, standard: claude-3-5-sonnet, flagship: claude-3-7-sonnet).

Layer 0.5: Dynamic Skill Router: Pre-indexes all 240+ skills into ~/.token-stack/skills-cache.json and enables Dual-Scope (Internal vs Harness) routing.

Layer 1: Code Topology Engine: Validates AST graph extractor & Graphify integration.

Layer 1.5: Data Lens & Columnar Engine: Probes ClickHouse HTTP port 8123 and provisions DuckDB / Zero-Row Stream Shield fallback.

Layer 2 & 3: Ponytail & Caveman: Safely enables caveman@caveman and ponytail@ponytail plugins in profile settings.json.

Layer 4: RTK CLI Output Filter: Validates RTK binary or provisions terminal log filter shim.

Layer 5-7: In-Flight Governors: Validates Turn Folding (5-turn Epoch Freezing), CoT Governor, and Loop Breaker circuit breaker.

Layer 8: Headroom Context Proxy: Checks daemon executables, upstream mappings (9284), and profile ports (8787+).

Layer 9 & 10: MemoraX & OpenViking: Creates episodic memory workspace ~/.token-stack/memory/ and context DB directories.

Global CLI: Registers global token-stack command and CMD wrapper in PATH.

🚀 Usage

Step 1: Preview Actions (Dry-Run)

token-stack setup

Step 2: Apply Configuration

token-stack setup -Apply

Step 3: Verify Probes

token-stack doctor

Pass -ProfileDirectory <path> to target a custom Claude/Codex profile directory.

name	token-stack:setup
description	Automated setup, configuration, and provisioning engine across all 14 layers of the Token & Context Engine. Use for initial setup, component repair, and global CLI registration.
user-invocable	true