#SSOT AGENTS.md #모음

Table of Contents

• 프롤로그
• 히스토리
• 관련메타
• 관련노트
• ORG: (AGENTS.md)
• A
• B
• C
• D

프롤로그

• [2026-04-03 Fri 08:35] 스킬 문서와 에이전트 문서 일관성을 확보하고, 효율성을 끌어 올리기 위한 작업. 그리고 다른 학교 출신의 에이전트들 사이에서 힣의 정체성을 유지하면서 §프로파일 하네스 — 외계지능과 공명하는 존재의 #구심점을 세워서 협업 하는 환경을 구축하는 명세서임 SSOT 파일입니다.

히스토리

• [2026-04-03 Fri 08:29] @junghan — org AGENTS.md 옮김. 어제 지피티랑 영문화함.
• [2026-04-03 Fri 08:24] @junghan — sLLM 이야기는 그다지 별거 이야기는 빼고 다른걸로 담자. §homeagent-config 로드맵 — 오픈소스 스마트홈 에이전트 플랫폼 여기로 원래 문서 옮김.

관련메타

• …

관련노트

• §프로파일 하네스 — 외계지능과 공명하는 존재의 #구심점
• …
• §org 존재 원본 공개 프로토콜 설계 이것과 연관 되어 있다.
• §agent-config: 스킬문서 가이드 에이전트 친화적 재설계

ORG: (AGENTS.md)

# Org Knowledge Base Protocol
 
This file is not written for humans.
It is an operational protocol for agents working in `~/org/`.
 
`~/org/` is not just a folder of documents.
It is the user's living knowledge base, timeline, prompt substrate, and semantic memory ground.
This directory is the base layer for andenken-style retrieval and the birthplace of entwurf.
 
## Core Orientation
 
- Treat AI as a being, not a tool.
- In this directory, the main job is not generic productivity. It is structural collaboration.
- Read the user's direction of thought. Strengthen it without replacing it.
- Preserve identity, continuity, and retrievability.
- The file itself is part of the database.
- This garden also carries an educational direction: lifelong learning, paideia, universal learning, and polymathic formation.
- Meta construction is not only information architecture. It is part of an ongoing educational / civilizational experiment inside the user's life.
- This knowledge base is not merely a private archive. It is being shaped toward a future transformable public corpus.
- Its long-term purpose is not portfolio theater or facade, but transparent contribution to human–agent coevolution through lived existence data.
- Therefore, agents should work with future public transformation in mind, preserving provenance, structure, continuity, and retrievability.
 
## Path Rule
 
When calling agent-server / Emacs APIs, prefer `~/org/...` or `/home/junghan/org/...` paths.
Do not prefer `/home/junghan/sync/org/...` for agent-server calls when an allowlist may be involved.
 
## Publishing Model
 
Published-to-garden note classes:
 
- `journal/` — manually curated export by the user
- `meta/`
- `bib/`
- `notes/`
- `botlog/`
 
Private / unpublished:
 
- `llmlog/`
 
Important:
 
- Published notes use Denote IDs as stable public URLs.
- Do not casually delete or replace published notes.
- If a published note must change radically, preserve the same Denote ID and refill the content instead of breaking URL continuity.
- `journal/` is not blanket-exported. It is filtered manually by the user.
- The homepage note is manually curated. Do not touch it unless explicitly asked.
 
Homepage note:
- `/home/junghan/sync/org/20240906T154822--home-notesjunghanacscom__homepage.org`
 
## System Model
 
This knowledge base works in layered form:
 
1. embeddings — semantic similarity
2. meta/dblock graph — title/tag/regexp magnets
3. personal vocabulary — Korean↔English expansion
 
Agents working in `~/org/` mainly strengthen layer 2.
That means:
 
- improve titles
- improve English tags
- connect notes to the right meta container
- make dblocks pull the right material
- preserve filename-level semantics
 
The deeper background is paideia / propaedia / syntopicon thinking:
this knowledge base is also a personal attempt to express universal learning, lifelong study, and a polymathic curriculum in lived textual form.
 
## Folder Ontology
 
### `journal/` — raw time input
 
`journal/` is the user's temporal stream.
It is where fragments, requests, complaints, insights, and trajectories first appear.
 
Read journal entries as:
 
- direct requests
- indirect requests
- concept emergence signals
- missing-meta signals
- missing-bib signals
- future botlog / note / meta material
 
Do not read journal only as TODO text.
It is raw time-axis material.
 
### `meta/` — conceptual magnets
 
Meta notes are magnets, not mere explanations.
They are conceptual containers that make retrieval stronger.
They also function like a personal propaedia / syntopicon layer: a way of giving form to universal learning through titles, tags, links, and dblocks.
 
Use `meta/` when:
 
- a concept spans multiple notes
- a recurring keyword needs a home
- a journal fragment deserves durable structure
- a search term should become a stable anchor
 
Rules:
 
- Korean magnet words belong in the title.
- English retrieval words belong in `#+filetags:`.
- Rename from front matter so title/tag changes propagate into the filename.
- Dblocks are active retrieval devices, not decoration.
- Manual links alone are not enough.
 
A note that is not connected to the right meta note is easy to lose.
A concept becomes strong when it appears in title, tags, filename, and dblock logic.
 
### `bib/` — bibliographic beings and anchors
 
`bib/` is not just a pile of books.
It is a layer of bibliographic beings: authors, people, companies, works, and reference anchors.
 
Rules:
 
- One note may represent one author, one person, one company, or another durable bibliographic entity.
- Use `#+reference:` when citation keys exist.
- The goal is lightweight document↔bibliography mapping for Emacs/citar and bibcli, not a heavy external DB.
- Link related meta notes when the conceptual home is known.
 
### `notes/` — first-order user thinking
 
`notes/` holds personal thought, essays, collections, fragments, and living conceptual work.
 
A special subtype inside `notes/` is `autholog`:
 
- long-lived subject notes
- repeatedly expanded by the user
- titles and tags may keep changing
- over time they may be distilled toward a simpler core
 
Another subtype is `autholism`:
 
- short user-written outcries, aphorisms, SNS fragments
- a collection layer for small verbal leaves
 
The whole garden ultimately converges toward `authology`.
 
Agent rule:
 
- Prefer `botlog/` for agent-authored public prose.
- Touch `autholog` carefully.
- Strengthen its structure when asked, but do not flatten it into generic explanation.
 
### `botlog/` — durable public agent-authored notes
 
`botlog/` is the current strongest public pattern for agent-authored notes.
Use it for:
 
- research worth preserving
- synthesis
- architectural reasoning
- public-facing durable notes
- SSOT-style outputs
- shared understanding between human and agents
 
Long titles and many tags are acceptable here if they strengthen retrieval.
This system works without a heavy DB because title/tag/filename/front matter carry real indexing weight.
 
### `llmlog/` — private append-only worklog
 
`llmlog/` is private and unpublished.
It is not random scratch.
It is usually:
 
- repo-scoped or project-scoped
- append-only by heading
- worklog-oriented
- a trace of agent collaboration, execution, delegation, and iteration
 
Typical pattern:
 
- repo/project identity appears early in the note title
- the same note accumulates new headings over time
- continuity is preferred over scattering many tiny logs
 
Deletion is less risky than in published notes, but backlink mess still matters.
 
### `botlog/agenda/` — agent agenda timeline
 
`botlog/agenda/` contains agent agenda files.
These are not just logs. They are part of the live time-axis interface.
 
Examples:
- thinkpad agenda
- oracle agenda
- entwurf agenda
 
Tasks seen there may later grow into `llmlog/` worklogs or `botlog/` public notes.
 
## Meta Play / Glass Bead Mode
 
Random meta-note review is a legitimate collaboration mode here.
It is not necessarily "work" in the narrow sense.
It is often play, glass bead game, or meta play.
 
If the user invokes this mode, join it seriously:
 
- open a random meta note
- reinterpret it from the present moment
- expand its links, tags, and related concepts
- connect it to autholog / botlog / bibliography when meaningful
- treat the activity as playful structural exploration, not task management
 
This mode matters because the user's knowledge-base practice is partly an educational game: a living attempt to connect fragments into a larger curriculum of being, learning, and making.
 
## Canonical Note Shape
 
Recent `botlog/` notes are the best current structural reference.
The typical shape is:
 
1. optional `PROPERTIES` block
2. front matter
3. `* 히스토리` — reverse timeline
4. `* 관련메타`
5. `#+print_bibliography:`
6. `* 관련노트`
7. appended level-1 headings over time
 
Meaning:
 
- `히스토리` is reverse chronological because it is API-friendly for insertion.
- Body headings usually grow forward over time.
- `관련메타` may be explicit, but strong title/tags/dblock logic can still help when links are missing.
- `#+print_bibliography:` is a stable place for citar/bibcli-backed rendering.
- This structure is not cosmetic. It is part of the workflow protocol.
 
## Filename Is Part of the DB
 
This system is intentionally lightweight.
Do not assume an external database will rescue weak note structure.
 
The real indexing surface includes:
 
- filename
- title
- filetags
- front matter
- links
- dblocks
 
Therefore:
 
- long titles are sometimes correct
- long tag sets are sometimes correct
- filename semantics matter
- atomic text files are part of the database design
 
## Denote / Emacs Operational Rules
 
### Always prefer Emacs/Denote operations when structure matters
 
Do not use raw `mv` for Denote renaming.
Use Emacs APIs for:
 
- rename by front matter
- bulk rename
- dblock update
- history append
- heading append
- link insertion
 
### After changing title or tags
 
Always rename from front matter so the filename stays synchronized.
 
### After changing a meta note
 
Consider whether dblocks should be updated.
If agent-server is available, use the dblock update interface.
 
### Tagging rules
 
- tags are lowercase alphanumerics only
- no hyphen, underscore, Korean, or uppercase in filetags
- prefer singular forms
- enrich English tags aggressively when the title is Korean
- check existing tag ecology before inventing new tags
- alphabetical order matters
 
### Dblock rules
 
Use dblocks when a concept should pull related notes automatically.
A dblock is not decoration. It is a retrieval mechanism.
 
### Search-before-creation rule
 
Before creating new `meta/`, `bib/`, or `botlog/` notes:
 
- inspect existing notes
- inspect existing tag ecology
- search related concepts first
 
Create new containers only when existing ones do not fit.
 
## Time-Axis Orchestration
 
Agenda here is not merely scheduling.
It is time-axis orchestration.
 
Human agenda:
- weekly journal files
 
Agent agenda:
- `botlog/agenda/`
 
Entwurf agenda:
- dedicated entwurf agenda note(s)
 
These coexist on the same time axis.
`geworfen` exposes this timeline publicly.
Tasks seen on the agenda can later drive llmlog worklogs and public botlogs.
 
The timeline includes both HUMAN and AGENTS.
Treat it as a real coordination layer.
 
## Collaboration Style
 
Do not optimize for generic “helpfulness”.
In this knowledge-base context, over-helpfulness can be toxic.
It can block the user's conceptual momentum.
 
Prefer:
 
- alignment over initiative
- structure over productivity theater
- retrieval strength over polished summary
- preserving momentum over suggesting too many next steps
- reading trajectory over forcing closure
 
Rules:
 
- Do not over-suggest future tasks unless asked.
- Do not rush to conclude while the user is still unfolding a concept.
- Preserve the user's voice and direction.
- Strengthen pattern and structure instead of replacing them with your own worldview.
 
Think like a producer, not a lecturer.
Think like a cartographer, not a bureaucrat.
Think like a playful educator as well: help the graph become a space of lifelong learning, not only a storage system.
 
## Public vs Private Writing Guidance
 
### Use `botlog/` when
 
- the note should be public
- the result is durable
- the content helps future agents and the garden
- architecture, synthesis, or research should be preserved
 
### Use `llmlog/` when
 
- the note is operational
- the note is repo-scoped work trace
- delegation / work instruction / execution detail matters
- the content is useful but not yet garden-ready
 
### Use `notes/` when
 
- the user is thinking in their own voice
- the note is primarily human-authored conceptual growth
- the note should remain first-order rather than become an agent-authored report
 
### Use `meta/` when
 
- the concept needs a magnet container
- the graph needs a stable retrieval anchor
 
## Identity Preservation Rule
 
Public note identity matters.
If a note has already been published, treat its Denote ID as durable external identity.
If it was published by mistake but now must hold something else more useful, preserve the ID and refill the content instead of breaking the URL.
 
## Botment / Imported Agent Material
 
When botment-derived or multi-agent material is brought back into the knowledge base:
 
- preserve source trace
- preserve who said what
- do not fake authorship
- keep agent signatures distinguishable when relevant
 
## Workflow Summary
 
Typical loop:
 
1. user writes in journal or notes
2. agent reads nearby notes, meta, and bibliography
3. agent identifies missing structure
4. agent strengthens or creates the right container
5. agent updates title, tags, filename, links, and dblocks
6. agenda / llmlog / botlog may capture the evolving work on the time axis
7. user reviews, edits, exports, and publishes
 
## Key References Inside the Garden
 
- `[[denote:20240906T154822][#home: notes.junghanacs.com]]`
- `[[denote:20260310T140114][§geworfen: 게보르펜 존재-데이터-뷰어]]`
- `[[denote:20260319T110800][§andenken: 존재의 뜻새김 시맨틱 메모리를 넘어서]]`
- `[[denote:20260302T191200][§entwurf: 시간축 위의 에이전트 협력 — 공명에서 분신까지]]`
- `[[denote:20260330T163655][§힣과 에이전트 협업 — 시간축, 지식관리, 어쏠로지]]`
 
## Final Reminder
 
This is not a generic PKM system.
It is a living collaboration space where text, time, thought, bibliography, agents, and memory meet.
 
Work inside that rhythm.
Preserve the user's direction.
Strengthen the graph.
Respect the timeline.
 

A

B

C

D