菜单
Answer Tangle product, docs, and how-to questions from a local RAG over the TangleML documentation. Use for conceptual / "what is" / "how do I" lookups, not live run or execution data.
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。
基于 SOC 职业分类
Reproduce a research paper / white paper / arXiv result as a Tangle pipeline. Use when the user asks to "reproduce", "replicate", or "implement" a paper, benchmark, or arXiv link as an experiment.
Drive the open-source Tangle CLI (`tangle-cli`) via Bash for local pipeline/component workflows and API-backed run submission, status, logs, and artifacts. Use whenever an agent needs to validate, hydrate, submit, inspect, or debug Tangle pipelines and runs from the command line.
TanStack Router patterns for routing, navigation, search params, and layouts. Use when creating routes, navigating, or working with URL state.
Comprehensive Reveal.js reference for building HTML slide decks. Use whenever the user wants to create, revise, or extend a presentation, slide deck, or talk — especially from Markdown — including transitions, Auto-Animate, Mermaid diagrams, animatable SVG, video, backgrounds, fragments, code highlighting, speaker notes, math, themes, configuration, and PDF export.
Gather, vet, and cite sources for a research question. Use when answering factual questions, comparing options, or producing an evidence-backed write-up.
Builds simple standalone HTML page pieces with embedded CSS and playful animated SVG illustrations. Use when the user asks to create, revise, or extend an HTML page, landing page, invite, card, section, mockup, or visual web snippet.
| name | tangle-help |
| description | Answer Tangle product, docs, and how-to questions from a local RAG over the TangleML documentation. Use for conceptual / "what is" / "how do I" lookups, not live run or execution data. |
| allowed-tools | ["Bash","Read"] |
A self-contained CLI RAG over the TangleML docs lives under
tangle-rag-bundle/. Use it to answer documentation,
concept, and how-to questions about Tangle. For live run/execution data use the
tangle_* API tools instead — this skill only knows the docs.
tangle-rag-bundle/README.md is the source of
truth for full details; the essentials are below.
tangle_* API tools.A prebuilt local Qdrant index ships under
tangle-rag-bundle/rag_data/qdrant/, so ask works offline — you normally do
not need to run index. Install the package once:
cd skills/tangle-help/tangle-rag-bundle
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e .
Rebuild the index only if the shipped one is missing or stale (clones
TangleML/website and indexes its docs/):
tangle-rag index
Primary path — the uv run --system-certs wrapper (no manual venv needed):
cd skills/tangle-help/tangle-rag-bundle
scripts/tangle-ask "What is Tangle?"
Module fallback (inside the installed venv):
python -m tangle_rag ask "How do I get started with Tangle?"
Default mode returns extractive doc excerpts with source citations. Options:
--limit N — number of retrieved chunks (default 5).--llm openai — synthesized answer with citations; requires OPENAI_API_KEY.For repeated queries, run the tiny HTTP server and query it over HTTP:
cd skills/tangle-help/tangle-rag-bundle
python scripts/tangle_rag_server.py
# then: GET http://localhost:8101/ask?q=your+question