Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/asaf-dahan/super-skill

A portable, AI-native intelligence layer that gives any language model full context, methodology, and operational capability over a defined domain.
https://github.com/asaf-dahan/super-skill

agent-skills ai-agent claude-code domain-experts knowledge-management llm notebooklm-py super-skill

Last synced: 2 months ago
JSON representation

A portable, AI-native intelligence layer that gives any language model full context, methodology, and operational capability over a defined domain.

Awesome Lists containing this project

README 

          
# Super Skill



Super Skill is an open framework for creating private domain experts

for you and your agents.



Each Super Skill is an intelligence layer: a self-growing knowledge base

that your agents read before every task, and a living view of your domain

expert's brain through NotebookLM -- so you can understand, audit,

and improve it over time.



Multiple agents. One source of truth. Always current.



---



## What This Means in Practice



Your agents stop starting from zero.



Every session, Claude Code reads your Super Skill before touching anything.

It knows what tools you run, what decisions you have made, what is waiting

for your approval, and what must never be re-litigated. It works from

verified current state, not assumptions.



You stop explaining yourself in every conversation.



NotebookLM holds the growing knowledge base for each domain.

You can listen to an audio overview of your own decisions,

take a quiz on your own architecture, and see a mind map of how

everything connects. As you learn, you improve the Super Skill.

As the Super Skill improves, your agents become more accurate.



---



## Skill vs Super Skill



| | Skill | Super Skill |

|---|---|---|

| What it is | A file | A living knowledge architecture |

| Knows how to | Do a task | Understand a full domain |

| Self-growing | No | Yes |

| Detects drift | No | Yes |

| Teaches the agent | Yes | Yes |

| Teaches you | No | Yes -- audio, quiz, mind map |

| Works with | Claude Code | Any agent, any model |

| Time to activate | 1 minute | Under 15 minutes |



---



## When to Create a Super Skill



Ask three questions before forking this template:



1. Does this domain change externally without your control?

   Tools release updates. APIs change. Pricing shifts.



2. Are there decisions that must never be re-litigated?

   Architectural choices, rejected alternatives, settled patterns.



3. Does your agent need verified current state before every task?

   Which plan, which version, which modules are active right now.



Two or more YES answers: create a Super Skill.

Fewer than two: a SKILL.md file is sufficient.



---



## How It Works



Prompt 1 runs in three stages. Paste the entire block into Claude Code.



```

Clone the template

        |

Open Claude Code in the folder

        |

Paste Prompt 1 from ONBOARDING.md

        |

Stage 1: Domain Layer Generation

  Claude Code scans your environment

  Claude Code asks at most 3 questions

  Claude Code generates all 10 layer files

        |

Stage 2: Expert Council Research

  Claude Code proposes 10 real domain experts

  You choose who joins your council

        |

Stage 3: Expert Profile Generation

  Claude Code generates expert profiles

  Claude Code generates experts/COUNCIL.md with debates

        |

Claude Code creates or updates ~/.claude/CLAUDE.md

(the global router that points to this Super Skill)

        |

Every future session:

Claude Code reads the router

Loads SUMMARY.md from the right Super Skill

Starts with full context in under 2 minutes

```



---



## The Layer Architecture



Every Super Skill contains the same 10 files regardless of domain.

The structure is identical. The content is yours.



```

SUMMARY.md          session entry point -- 80 lines, always current

SKILL.md            agent skills format entry point

CLAUDE.md           claude code operating instructions

AGENTS.md           instructions for any other agent

CONTEXT.md          layer 0: identity, principles, scope

DOMAIN_MAP.md       layer 1: domain structure and sub-domain relationships

CURRENT_STATE.md    layer 2: verified current state

EVALUATION.md       layer 3: framework for evaluating anything new

DECISIONS.md        layer 4: decisions log and reasoning

MONITORING.md       layer 5: drift detection sources

LEARNING.md         layer 6: notebooklm notebook plan

PENDING.md          layer 7: approval queue

experts/COUNCIL.md  layer X: expert council and debates

experts/[name].md   individual expert profiles

notebooks/          notebooklm-ready learning files

scripts/            automation: feed, generate, check, sync, update_summary

changelog/          monthly detected changes

```



---



## The Operating Principle



The model proposes.

You decide.

The Super Skill records.

The system executes.



No layer file changes without your approval.

Every proposed change goes to PENDING.md first.

You review, you decide, then and only then it is recorded.



---



## NotebookLM -- The Learning Layer



NotebookLM is optional but transforms what a Super Skill can do.



Without NotebookLM: your agents read Markdown files from the repository.

Full context, zero re-explanation, every session.



With NotebookLM: the knowledge base grows beyond static files.

You and your agents can add research, documents, and new sources.

Claude proposes additions with your approval. You add directly in the notebook.

The notebook generates audio overviews, quizzes, and mind maps from your

own domain knowledge. Token usage drops significantly because Claude queries

the notebook instead of loading full files.



Connect it in Step 4 of ONBOARDING.md. Skip it and add it later if you prefer.



---



## Expert Council



During activation, Claude Code researches your domain and proposes

10 real, leading experts with verifiable published work. You choose

who joins your council. Each expert is assigned to the layer they

contribute most to and given a full profile with methodology,

frameworks, red lines, and five questions they would ask you.



When experts disagree, the disagreement is recorded as a debate in

experts/COUNCIL.md. Claude Code surfaces expert perspectives only

when genuine friction exists -- not on routine tasks.



If an expert's position conflicts with a decision in DECISIONS.md,

it goes to PENDING.md as an expert challenge for your review.



---



## Domain Types



Super Skill works for any domain where knowledge matters and decisions accumulate.



Technical: software infrastructure, database architecture, security frameworks,

DevOps pipelines, product methodology, AI agent systems.



Professional: investment portfolio, legal review, medical documentation,

marketing strategy, real estate, financial modeling, operations.



Personal: home garden, fitness, nutrition, home renovation, language learning,

travel planning, personal finance.



Craft: photography, woodworking, wine, cooking techniques, academic research.



If you can describe it, a Super Skill can master it.



---



## Get Started



Clone this repo. Open Claude Code. Paste Prompt 1 from ONBOARDING.md.



  git clone https://github.com/Asaf-Dahan/super-skill my-super-skill-[domain]

  cd my-super-skill-[domain]

  claude .



Your Super Skill will be live in under 15 minutes.



---



## Version



Super Skill v2.1.0



## License



MIT License -- (c) 2026 Gitit Inc - AI Architecture



Fork it. Build your own. Publish your Super Skills.



---



## Important Notes



notebooklm-py is an unofficial library that uses undocumented Google APIs.

It may change without notice. Suitable for personal and internal use.

Not recommended for production systems serving external users.



Keep domain-specific Super Skills in private repositories.

Never commit API keys or tokens to any layer file.

Use .env for all sensitive values.



All agent actions follow one rule:

The model proposes. You decide. Nothing changes without your approval.



---



## Created By



Super Skill was built by Asaf Dahan,

AI Solutions Architect at Gitit Inc.



Website: ai.asafid.com

Company: Gitit Inc

GitHub: github.com/Asaf-Dahan