https://github.com/solitonschooler/research-continuity-skills

Research Continuity Skills - Released on GitHub

 Please comment if this git has any special usage.

https://github.com/solitonschooler/research-continuity-skills

  

Yes — this repo has several distinctive / “special” uses beyond a typical README + code dump.

Primary intended use: portable Agent Skills for research continuity

It ships two LLM-native Agent Skills (compatible with the agentskills.io format):

Skill	Purpose
research-kernel-builder	Consolidate a messy corpus (papers, notes, chat exports, design docs, agent logs) into a source-grounded, versioned, recoverable “research state” (Kernel). Supports optional verify / challenge / repair / develop / experiment modes.
research-archaeologist	Reconstruct and audit the history of research or technical decisions — where ideas came from, how errors propagated, what was abandoned, unsupported assumptions, etc. Does not require a pre-existing Kernel.

Key design points that make them special:

• Evidence discipline is non-negotiable: distinguishes source assertions from agent reconstructions, keeps negative results and residuals, tracks lineage and corrections, and forbids claiming verification that did not happen.
• No runtime dependencies: pure declarative skills (SKILL.md + references + schema). No database, Python package, or API key required by the skills themselves.
• Effort profiles (SCAN, COMPILE, VERIFY_REPAIR, DEVELOP, EXPERIMENT, ITERATIVE_FULL) control scope; you can stop at any time and get a resumable checkpoint.
• Shared JSON Schema + evidence contract so outputs stay machine-readable and handoff-friendly across sessions/agents.

This is aimed at long-running, multi-document, AI-assisted research that normally loses provenance and continuity.

Secondary / “hidden” special use: SMFT research specimen + toy AGI architecture

The repo is also a public mirror of a large original research corpus (325 authored documents under data/smft/src/) plus selected kernels (v8.6 Light + historical excerpts).

There is an explicit easter egg (EASTER_EGG_SMFT_AGI.md) that frames the included SMFT / World-Formation Kernel as a cross-layer toy architecture for studying adaptive agents. It outlines a 6-layer candidate stack (micro-dynamics → observer/world declaration → P8D typed interfaces → functional ports → G/A/C/S/R governance → research ledger) with conditional mathematics, negative results, and proposed falsifiable experiments. The author is careful to label it as exploratory engineering, not implemented AGI or proven Transformer mechanisms.

In short, the same Skills that were forged during ~86 rounds of kernelizing this corpus can now be turned back on the corpus itself (or on any other research history).

Other notable traits

• Strong emphasis on adversarial robustness (synthetic adversarial mini-corpus, acceptance gates, instruction-injection resistance).
• Explicitly experimental / pre-license: the README states a license has not yet been selected, so public visibility ≠ unrestricted reuse rights.
• Zero stars / very early stage at the time of inspection; maturity claims are modest and self-documented.

Bottom line: The “special” usage is as installable, evidence-strict Agent Skills for reconstructable research continuity, with a built-in large real-world specimen (SMFT) that also doubles as a speculative but carefully caveated multi-layer agent architecture playground. It’s not a conventional software library — it’s closer to a methodological toolkit + research archive designed to be operated by LLMs.

 

 

Is its functionality looks quite "unique"? Or it is some kind of skills that been invented every week - if not every day?

 

Not unique in the broad category, but meaningfully distinctive in its specific focus and rigor.

The broader landscape (very crowded)

Agent Skills (the SKILL.md portable format) exploded after Anthropic formalized the open standard in late 2025. By mid-to-late 2026 there are surveys of the field, marketplaces with thousands of skills, and new ones appearing constantly. Research-related skills are especially common:

• End-to-end research pipelines (idea → literature → experiment → paper writing)
• Literature search / evidence extraction / citation verification
• Academic writing, plotting, peer-review assistance
• Domain-specific scientific procedures (genomics, cheminformatics, etc.)
• General session/memory continuity skills (remember facts, wake-up briefs, project state snapshots)
• Knowledge-graph / claim-evidence systems and research-state kernels

So “skills that help agents do research or keep continuity” are invented frequently — multiple times a week is realistic.

What makes this one stand out

research-continuity-skills is not another literature-review or paper-writing skill. Its core is narrower and deeper:

• Turning messy, multi-version, multi-author corpora (papers + notes + chat logs + agent histories) into a source-grounded, reconstructable research state (“Kernel”).
• Explicit Research Archaeologist mode that reconstructs history — where ideas originated, how errors propagated, what was abandoned, unsupported assumptions — without requiring a prior Kernel.
• Very strict evidence contract: source assertion ≠ agent formalization ≠ verified result; mandatory tracking of negative results, residuals, lineage, and “INSUFFICIENT_EVIDENCE”; no claiming verification that didn’t happen.
• Effort profiles + stop-anytime checkpoints designed for long-running, multi-session work.
• Comes with a large real specimen (325 SMFT documents + evolved kernels) that was itself produced by the methods the skills encode.

Closest relatives exist (e.g., academic research kernels with claim-evidence graphs, decision ledgers, or provenance tracking), but the combination of pure declarative Agent Skills + research-state kernelization + historical archaeology + adversarial evidence discipline + a full public research corpus as testbed is uncommon. Most other research skills focus on producing new research or session memory; this one is optimized for recovering, auditing, and continuing existing research without losing provenance or honesty.

Bottom line

• Category level: Ordinary — part of a very active wave of agent skills.
• Specific functionality and philosophy: Distinctive. It feels more like a carefully engineered methodological toolkit forged from a real multi-month research effort than a quickly generated “do research better” skill.

If you’re evaluating it for actual use, the interesting question is less “is this the only one?” and more “does its evidence discipline and archaeology focus match the kind of messy longitudinal research you actually do?”

 

 

 

© 2026 Danny Yeung. All rights reserved. 版权所有 不得转载

 

Disclaimer

This book is the product of a collaboration between the author and OpenAI's GPT 5.6, Google AI, Gemini 3.X, NoteBookLM, X's Grok, Claude' Sonnet 5 language model. While every effort has been made to ensure accuracy, clarity, and insight, the content is generated with the assistance of artificial intelligence and may contain factual, interpretive, or mathematical errors. Readers are encouraged to approach the ideas with critical thinking and to consult primary scientific literature where appropriate.

This work is speculative, interdisciplinary, and exploratory in nature. It bridges metaphysics, physics, and organizational theory to propose a novel conceptual framework—not a definitive scientific theory. As such, it invites dialogue, challenge, and refinement.

I am merely a midwife of knowledge.