双循环进化：内部反思(P0) + 外部吸收(P1)。Cross-project absorption methodology — multi-round cross-project comparison, active project tracking, self-expanding keyword discovery. 动灵驱动吸收(Entelechy-Driven Absorption v4.3).

2026-06-225

Core Research Stack — 核心科研栈

Purpose

Navigation index for the core research stack: 7 cognitive atoms, paper pipeline, research methodology, and AI/ML tools.

Skills in this Layer (88 total)

academic-diagram: Directory index for academic-diagram: academic-diagram

adhd-eye-tracking-review: Directory index for adhd-eye-tracking-review: adhd-eye-tracking-review

akne-knowledge-manager: AKNE 知识管理系统的双向整合审计、Synthos-桥接诊断、知识流分析、内容级审计。与 akne-maintenance 不同，专注两系统间的连接质量及知识内容质量而非内部运维。

akne-maintenance: 维护个人知识库系统AKNE — 图谱诊断、修复、向量填充、源文件覆盖、Wiki清理、自动进化守护。覆盖AKNE仓库的完整运维生命周期。

architecture-diagram: Directory index for architecture-diagram: architecture-diagram

argument-expression: Directory index for argument-expression: argument-expression

arxiv: arXiv论文搜索 — 按关键词/作者/类别/ID检索。支持Tor SOCKS代理访问。

association-discovery: Directory index for association-discovery: association-discovery

audiocraft: Comprehensive guide to using Meta's AudioCraft for text-to-music and text-to-audio generation with MusicGen, AudioGen, and EnCodec.

autonomous-execution-threshold: ≥80%置信度 = 闭嘴执行。 不输出推测文案、不给选项、不喊"开始自主执行"口号。用户看到的是执行结果，不是选择题。

bib-integrity-audit: Audit .bib reference files across a paper library for:

biomechanical-regulation-ode: Class-level skill for building computational dynamical models of physiological regulation systems using 2-ODE systems with PINN training. Covers model formulation, bifurcation analysis, Sobol sensitivity, and ablation studies.

biorxiv: Directory index for biorxiv: biorxiv

bppv-expert: Structured BPPV (Benign Paroxysmal Positional Vertigo) medical knowledge version: 1.0.0 extracted from AKNE knowledge graph. Covers diagnosis techniques (Dix-Hallpike, supine head flexion test), repositioning maneuvers (Epley, Gufoni, Semont, Barbecue, roll-over), canalith conversion mechanisms, 3D biomechanical simulation, and clinical decision workflows. Source: AKNE wiki (126 nodes, 137 edges, proven correctness via falsification testing).

citation-bib-crossref: Scan paper directories for mismatches between \\cite{key} calls in .tex files and @type{key} entries in .bib files. Produces D8 (bib count) and D10a (match percentage) metrics, plus orphan/zombie classification.

citation-integrity-fix: ```python

cognitive-atom-architecture: Methodology for transforming operational skill sets into independent cognitive atoms with strict DAG dependencies, input/output contracts, and Synthos framework alignment. v4.0.0 syncretic framework: Eastern ontology (格物通理/取象通变/天人合一) + Western epistemology (经权度信/墨证求真/庄周观模) + 熵减律·生生之谓易 (ultimate purpose) + 大道至简 (cross-cutting razor).

comfyui: Directory index for comfyui: comfyui

competition-submission: End-to-end preparation of AI/tech/innovation competition submissions. Extract requirements, map to scoring criteria, generate documents, produce submission checklist. Covers medical AI, tech innovation, academic conferences, Chinese government grants.

conversation-to-memory: ⚡ P0 辅助技能。从会话中提取高价值信息，在2,200字符限制内最大化记忆ROI。动灵记忆三问(生长方向?框架维度?发酵潜力?)+宪法护栏(记忆不能覆写CONSTITUTION)+凝练压缩策略。防记忆膨胀同时保留高信号事实。每条记忆标注生长方向和发酵潜力。

crispdm-helix-experiment: CRISP-DM Helix methodology — strict CV-fold-isolated preprocessing for clinical ML experiments on public datasets. Generates real, traceable L0.5-compliant data.

data-driven-hypothesis: 从公开数据出发→数据探索→文献调研→发现gap→提出可验证假设。与"先有假设再找数据"相反，适合公开数据集方向探索。

dspy: Use DSPy when you need to:

emerging-field-landscape-scan: Skill: emerging-field-landscape-scan

evolution: ⚡ P0 自进化引擎。Synthos evolution engine v2.20 — 四态决策+硬收敛+GEPA反射分析+自动基准+Pareto优化+外部吸收+教训学习+黄金验证+自扩关键词+漂移检测+渐进披露+Git即记忆。Hooks注入+置信度评分+并行Agent审计+会话上下文注入+Prompt Snippets。

experiment-recipes: ML训练配方与预设——架构选择、训练循环、优化器、调度器、混合精度、内存优化、调试。提炼自实战经验，非外部代码搬运。每个配方记录原理而非逐行代码。

falsification-validation: Systematic approach to validating AI agent skills through falsification

golden-test-methodology: Methodology for creating, maintaining, and evaluating golden test suites version: 1.0.0 across all skills. Defines: three-file golden structure (GOLDEN_SET.md + cases/ + expected/), weighted verification criteria, coverage scoring, and DIAGNOSE integration. Covers the systematic gap where golden coverage is the weakest dimension in a multi-skill system.

gradient-alignment-loss: Skill: gradient-alignment-loss

hcs-3wt-breast-cancer-diagnosis: HCS-3WT (Hybrid Cascade-Stacking Three-Way Triage) breast cancer diagnostic

healthcare-dataset-discovery: Public healthcare dataset discovery — known accessible sources, dead sources, and API patterns for medical AI research.

huggingface-hub: HuggingFace hf CLI: search/download/upload models, datasets.

hypothesis-generation: Directory index for hypothesis-generation: hypothesis-generation

inference: Directory index for inference — mlops/inference 模型推理服务与优化

journal-selection-medical-ai: Systematic methodology for evaluating and ranking SCI journals as publication

kg-bridge: Knowledge Graph — Agent Bridge. 将大型知识图谱接入 Agent 记忆层的方法论，覆盖查询分层、环境隔离、语义搜索增强、脚本化接口。

knowledge-acquisition: 多源学术论文检索：Semantic Scholar / PubMed / Crossref / OpenAlex / arXiv / bioRxiv。

knowledge-base-audit: Audit and maintain personal knowledge management systems (AKNE, NotebookLM,

knowledge-extraction: Directory index for knowledge-extraction: knowledge-extraction

latex-output: Directory index for latex-output: latex-output

literature-monitor: Directory index for literature-monitor: literature-monitor

llama-cpp: Use this skill for local GGUF inference, quant selection, or Hugging Face repo discovery for llama.cpp.

metacognition: 元认知 — 自主执行阈值、记忆增强、记忆优化系统。

models: Directory index for models — mlops/models 模型架构

nano-pdf: Edit PDFs using natural-language instructions. Point it at a page and describe what to change.

nature-paper2ppt: Nature-style Chinese PPTX from academic papers — argument-driven slide

nsfc-grant-audit: Directory index for nsfc-grant-audit: nsfc-grant-audit

obliteratus: 9 CLI methods, 28 analysis modules, 116 model presets across 5 compute tiers, tournament evaluation, and telemetry-driven recommendations.

ode-simulation-tuning: When building 2-ODE computational models for SCI papers, the simulation rarely passes all quality gates on first run. Metrics that commonly fail: R² (needs smooth fit), AUC (needs proper distribution comparison), ablation (needs dominant coupling), accuracy (needs clean transitions). This skill captures the systematic tuning methodology used across papers 90–140+.

openalex: Directory index for openalex: openalex

outlines: Use Outlines when you need to:

paper-citation-health: Scan all papers in outputs/papers/ for citation bibliographic health metrics D8 (bib entries) and D10a (cite-to-bib match %).

paper-cron-scan: Cron job 不是完整的论文管线执行，而是轻量级扫描：验证白空间稳定、发现新竞争、推进管线状态。

paper-pipeline: 主skill | SCI论文全流程编排器。v3.18.10新增Trap#42跨项目参考文献污染检测（Synthos Paper ID后缀/占位符键名/空条目/Prose提及无cite）。v3.18.9新增Trap#41 paper-queue.json幽灵条目逆方向。v3.18.5-8: D10a批量扫描+natbib盲区+注释过滤+路由修复。v3.18: Track A晋升协议。v3.16: 队列自愈+ABSOLUTE WHITE独立验证。v3.15: 轨道B四步工作流。

paper-quality-deep-review: 论文质量深度审查引擎 — 从文献下载→内容分析→研究空白验证→科学假设评估→解决方法评估→文献引用质量评分→综合评分。

paper-queue-audit: Directory index for paper-queue-audit: paper-queue-audit

paper-references-scanning: Scan paper library for citation health: D8 (bib entry count), D10a (cite-to-bib match rate), orphans, zombies. Class of tasks: LaTeX reference integrity auditing.

patent-disclosure: 该技能以Synthos仓库为主版本。Hermes镜像为查找索引，执行时请加载Synthos路径版本。

pdf-download-racing: 并行竞速PDF下载引擎 — curl_cffi TLS指纹绕过 + Sci-Hub域轮换 + LibGen + MedData。依赖 tools/paper-manager/src/。

pdf-to-md-notebooklm: PDF→Markdown→NotebookLM 全流程管线。支持批量上传、自动类型检测、大文件处理。

political-proposal: 参政议政提案全流程(民进/政协/人大) — 三段式: 基本情况→问题→对策。

project-experience-distillation: ⚡ 最高优先级技能。From project experience to reusable skill — extract workflow patterns, design principles, and pitfalls from completed project work, abstract them into general form, and formalize as SKILL.md. Also: philosophical implementation gap analysis to drive mechanism-level improvements. The reflexive learning engine of Synthos: self-evolution through self-observation.

pubmed: Deep PubMed/MEDLINE search via NCBI E-utilities — query construction, MeSH terms, batch retrieval, clinical query refinement.

quality: 质量保障 — 伪证验证、黄金测试、SCI论文质量评审。

quality-score-assignment: Paper satisfies current_step in steps_completed AND len(steps_completed) >= 8.

research: 直接调用子类别/技能名称即可。例如：arxiv、bib-integrity-audit、research-ideation。

research-ideation: 研究创意发散与认知引擎（RIF+CCF）。三层架构：Layer 1（10操作框架）→ 产出研究方向候选； Layer 2（8认知引擎）→

research-paper-search: 主skill | 多源论文检索+全文下载编排器。入口：Semantic Scholar (API Key), PubMed, OpenAlex, arXiv (Tor), Crossref。调用子skill: arxiv, pubmed, openalex。

research-skill-audit: Audit and enhance research skill coverage. Process for identifying gaps, testing existing skills, and creating/enhancing missing capabilities.

researcher-portrait: Directory index for researcher-portrait: researcher-portrait

scc-bppv-kinematics: Skill: scc-bppv-kinematics

sci-paper-quality-review: Directory index for sci-paper-quality-review: sci-paper-quality-review

sci-paper-standard-structure: Directory index for sci-paper-standard-structure: sci-paper-standard-structure

scientific-database-lookup: Directory index for scientific-database-lookup: scientific-database-lookup

segment-anything: Comprehensive guide to using Meta AI's Segment Anything Model for zero-shot image segmentation.

skill-absorption: 双循环进化：内部反思(P0) + 外部吸收(P1)。Cross-project absorption methodology — multi-round cross-project comparison, active project tracking, self-expanding keyword discovery. 动灵驱动吸收(Entelechy-Driven Absorption v4.3).

skill-integrity-audit: | 概念 | 文言 | 义 |

synthos: >

synthos-akne-bridge: Synthos 与 AKNE 之间双向桥接 — 论文目录规范化、技能连接、逆向边创建、Wiki 清理、自动守护重启、内容摘要注入、向量化补全。与 akne-maintenance（内部运维）和 akne-knowledge-manager（审计诊断）不同，本技能管具体的桥接操作。

system-bridging: 跨系统连接模式 — 两个独立系统（知识图谱/论文管线/技能库/监控系统）之间的双向桥接。覆盖连接协议、数据注入、反向查询、同步守护、重叠检测。

systematic-review: 系统综述与Meta分析工作流助手 — PRISMA流程、搜索策略设计、研究选择、质量评估、数据提取和综合支持。

task-router: Synthos系统入口。路由用户查询到正确的认知原子链或执行模式。四模式：标准链 / 探索循环 / 研究双循环 / 并行执行。 Agent-native执行，纯skill驱动零Python。

training: Directory index for training — mlops/training 模型训练与微调

v32-multi-direction-scan: Every cron run of autonomous-core-researcher after v31 API fix. Standardized pattern for scanning 5 rotation + 5 new directions per run.

viewpoint-verification: Directory index for viewpoint-verification: viewpoint-verification

vllm: Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.

writing: 写作辅助 — 引用完整性修复、LaTeX输出、政治提案、标准论文结构。

IO_CONTRACT

input: layer: str, query: str, context: dict — Layer name, search query, and context

output: skill_list: list[dict] — Filtered list of skills matching the query

对应原则：P3（人机分层 — 路由器负责路由，原子负责执行）