aistec-skills

🌐 中文 | English

AISTEC Skills

AI-driven Integrated Science & Technology Education Content Creation Framework

Overview

This repository contains three OpenClaw AgentSkills that power the AISTEC (AI-driven Integrated Science Education Content Creation Framework) system.

The framework integrates:

• Citation verification via multi-API cross-check (Semantic Scholar + arXiv + CrossRef + Google Scholar + CNKI)
• 13 verified pedagogical frameworks auto-matched by decision tree (5E, 7E, 4-Stage, PBL, Design Thinking, IBL, POE, REACT, LIA, UbD, GRRF, IDM, Guided Play)
• UDL × SDT integrated design — Universal Design for Learning fused with Self-Determination Theory (autonomy × competence × relatedness)
• SSI termination condition — automatic halt when topics involve scientific controversy, ethical decisions, or social policy

Skills

1. research-lineage

Research lineage intelligent engine. Applies bibliometric analysis methods informed by CiteSpace/VOSviewer (clustering quality, betweenness centrality, burst detection), combined with PRISMA-style screening records and CoE citation verification.

Key features:

• 6-round joint retrieval (web search + arXiv/Google Scholar via API + CNKI browser automation), plus supplementary verification rounds
• Citation graph construction (seed papers → 2-hop citation graph → elite pool via LLM scoring into Core + Adjacent tiers)
• CoE 6-step verification pipeline (A→B→C→D→E→F, internal workflow): graph building → API screening → bibliographic cross-check → multi-API cross-verification → LLM summary entailment → timeline consistency + upstream contamination detection
• 5-class hallucination detection (TF/PAC/IH/PH/SH)

2. science-edu-activity

Science education activity plan generator. Grounded in constructivist learning theory and inquiry-based learning, supporting UDL × SDT integrated design, ICAP cognitive engagement assessment, metacognition embedding, and visible-hidden line alignment checking.

Key features:

• Dual-format output — every activity plan includes both (1) a human-readable Markdown lesson plan and (2) a machine-parseable page-level executable scheme ready for digitization (OpenMAIC import, interactive classroom deployment, etc.)
• 4 page types in executable scheme: slide (SL), interactive (IA), quiz (QZ), pbl (PB) — each with learning contract, interaction design, assessment, scaffolding, pedagogy tags, narration, and timing
• 13 pedagogical frameworks auto-selected by decision tree (5E, 7E, 4-Stage, PBL, Design Thinking, IBL, POE, REACT, LIA, UbD, GRRF, IDM, Guided Play)
• Knowledge Graph integration — automatic prerequisite diagnosis via AISTEC Knowledge Graph before activity design
• UDL × SDT integrated design — fuses Universal Design for Learning with Self-Determination Theory so "multiple means of engagement" is backed by psychological need satisfaction
• ICAP cognitive engagement assessment — assigns P/A/C/I target levels per activity stage, designs upshifting strategies to ensure cognitive engagement progression
• Metacognition embedding — marks think-aloud prompts, reflection cards, and self-questioning checkpoints in the activity flow
• Visible-hidden line alignment check — ensures every "fun" activity segment maps to a clear learning objective
• Internal formative assessment feedback — auto-corrects plan without exposing evaluation rubrics
• Multi-dimensional internal feedback loop (scientific accuracy, educational effectiveness, inquiry process, learning experience, cognitive engagement, motivation design)
• SSI termination condition — halts generation for controversial ethical/policy topics
• CoE citation verification integrated into scientific background knowledge validation
• STEM/STEAM integration mapping, NGSS 3-dimensional alignment

3. science-edu-template-trainer

Template extraction and migration learning engine. Converts historical activity plans into standardized templates via 9-phase multi-dimensional quality analysis.

Key features:

• 9-phase pipeline: input ingestion & validation → individual analysis → content analysis & linguistic assessment → quality scoring & benchmarking → pattern aggregation → gap analysis & equity audit → template generation → integration with science-edu-activity → validation & delivery
• Pedagogical framework analysis (theory basis, learning objectives, assessment alignment)
• Fairness audit (cultural responsiveness, gender equity, accessibility, socioeconomic inclusion)
• Cost-benefit analysis (per-capita cost, time efficiency, educational ROI)
• Context-aware customization (museums, schools, outdoor programs, workshops)

Verified Frameworks

All pedagogical frameworks referenced in the skills are verified against real academic sources:

• 5E — Bybee et al. (2006)
• 7E — Eisenkraft (2003)
• 4-Stage — Museum/Exhibition inquiry model
• PBL — Krajcik & Blumenfeld (2006)
• Design Thinking — d.school / IDEO methodology
• IBL — Pedaste et al. (2015)
• POE/PEE — White & Gunstone (1992)
• REACT — Crawford (2001)
• LIA — PrimaryConnections (2024)
• UbD — Wiggins & McTighe (2005)
• GRRF — Fisher & Frey (2013)
• IDM — Grant et al. (2017)
• Guided Play — Hirsh-Pasek et al. (2016)

Installation

These are OpenClaw AgentSkills. Place each skill directory under your OpenClaw skills path (e.g. ~/.openclaw/workspace/skills/).

Requirements

• OpenClaw with kimi-search plugin
• Access to: Semantic Scholar API, arXiv API, CrossRef API, CNKI (via browser automation)
• For science-edu-template-trainer: pandoc (for Word/PDF conversion)

Platform Dependencies

Two skills contain tool calls specific to the Kimi Claw (Moonshot AI) platform. All pedagogical frameworks and design logic are platform-agnostic — only the tool invocation syntax needs adaptation when porting.

research-lineage — Heavy dependency (29 references in SKILL.md)

The entire CoE verification pipeline routes through four Kimi-specific tools:

Porting method: Rewrite only the tool routing table in SKILL.md (search strategy section), replacing each kimi_* call with your platform's equivalent. The CoE 6-step verification logic, hallucination detection, and citation graph construction are entirely platform-independent.

science-edu-activity — Light dependency (10 references)

Tool call examples appear in two places:

1. SKILL.md — API call examples for literature retrieval (7 references)
2. references/accuracy-assurance.md and references/knowledge-deconstruction.md — tool names in verification workflows (3 references)

Porting method: These are example-level references. Replace kimi_search with your search tool, kimi_datasource_call with your academic API wrapper, and kimi_fetch with your fetch tool. The 5-stage knowledge deconstruction, 13-framework decision tree, UDL × SDT design, and all pedagogical logic require zero changes.

No dependency

science-edu-template-trainer — fully platform-agnostic.

License

MIT