A personal AI-assisted creative workshop for worldbuilding, fiction systems, fanfiction/daydream simulation, research, concept development, prompt authoring, and image workflows.
Creative workflow · Creative systems · Central Hub · Prompt Authoring / AT · Choose your path · Releases · System context
This is the author's personal, non-commercial working environment, developed with ChatGPT and Codex. The fiction projects are the creative work, not sample data for an AI framework.
The repository began with ChatGPT Custom Instruction versioning. It now contains externalized fictional sources, creative methods, research and image workflows, alongside the LLM controls that support them. Routing, anti-drift controls, regression checks, and design history grew from actual use, observed failures, bounded changes, and feedback.
CREATIVE WORK = PURPOSE
CONTROL INFRASTRUCTURE = SUPPORT
CI means Custom Instructions, not Continuous Integration. Historical/internal paths and the current GitHub repository name remain in use. This is an independent personal project, not an official OpenAI project or a guarantee of ChatGPT behavior.
A fiction idea can go straight to a fiction project. Central Hub is available when shared reasoning, research, concept, prompt, or image capabilities are useful; it is not a required intake step.
IDEA ───────────────────────────→ independent fiction projects
└→ Central Hub when useful ←──→ FTH / The Kingdom / Fanfiction Engine
Valid flows include IDEA → FTH, IDEA → The Kingdom, and IDEA → Fanfiction Engine, as well as FTH → Central Hub → FTH, The Kingdom → Central Hub → The Kingdom, and Fanfiction Engine → Central Hub → Fanfiction Engine.
When work leaves a project for Central Hub, the Hub works on the supplied material without inheriting that project's canon authority. The result returns as working material. The receiving project's own sources, local logic, controls, and authorial review determine what the result means there and whether any part is accepted.
A transferable result may be re-abstracted into a new fiction idea and explored elsewhere. Its local implementation does not transfer automatically: each project evaluates the idea under its own logic and author decisions.
SHARED FICTION IDEA != SHARED IMPLEMENTATION
HUB OUTPUT != PROJECT ACCEPTANCE
LOCAL FICTION LOGIC > CROSS-FICTION ANALOGY
CENTRAL HUB != SEMANTIC PARENT OF FICTION PROJECTS
These independent creative systems keep their own canon, ontology, source authority, and local controls.
| Project | Repository entry | Interpretation |
|---|---|---|
| Project Feather Core (FTH) | Project Feather Core | Worldbuilding with its own sources and authority; AetherFire is a country inside FTH. |
| The Kingdom | The Kingdom | Its own world model and local rules; do not infer its architecture from another project. |
| Fanfiction Engine | Fanfiction Engine | Fanfiction/daydream simulation system; The Academy is its internal platform. Its operational domains can run independently or compose. |
Each project's documentation determines how to read its sources. Folder names, version numbers, archives, or resemblance to another project do not establish canon authority.
Central Hub dùng Architecture/Catalog 1.22 với bảy package và 23 Library controls trong repository. Người dùng đã thử thành công việc đọc và sử dụng folder Library lồng nhau trong runtime liên quan. Bộ 1.22 chưa được cài hoặc thử live. Central Hub. User Profile giữ shared non-module identity riêng. Các release distribution giữ tên lịch sử.
Central Hub is the shared general-purpose workspace for capabilities that should not belong to any one fiction project. It does not sit inside FTH, above the creative systems, or act as a shared fiction runtime or canon layer. A project may bring a bounded question or slice of context into Hub and return the result for local review; Hub may also start from a fiction idea before any project has adopted it.
Central Hub coordinates a mission: the user's current task. It keeps the overall objective while composing contributions with authority limited to their assigned portions. Ordinary chat, support, writing, or lightweight brainstorming may need no specialized package.
These are behavioral contracts for model-guided work, not an executable routing program. The topology below shows selection and composition, not a mandatory sequence through every package.
USER TASK
↓
HUB CI — operation/stage · truth/state · action authorization
↓
HUB ROUTER — scope · selection · dependency/read validation
↓
ROOT PACKAGE CATALOG
├─ HUB-PKG-TECHNICAL → HUB-TECH-ENTRY
├─ HUB-PKG-CONCEPT-DEVELOPMENT → HUB-CON-ENTRY
│ └─ local catalog
│ ├─ CREATIVE
│ ├─ BRAINSTORM
│ ├─ DEBATE / NAMING
│ └─ task-selected anti-drift controls
├─ HUB-PKG-AUTHORING → HUB-AUTH-ENTRY
├─ HUB-PKG-AIS → AIS pipeline (internal dependencies)
├─ HUB-PKG-IMAGET → HUB-IMAGET-ENTRY (internal dependencies)
├─ HUB-PKG-SHOPPING → HUB-SHOP-ENTRY
└─ HUB-PKG-FOOD → HUB-FOOD-ENTRY
↓
selected modules / package-local leaves / tool pipelines
↓
HUB COMPOSITION → WORKING RESULT
Hub CI owns the requested operation/stage, truth and state discipline, action authorization, general verification need, and overall interaction. Hub Router scopes task portions, selects root packages, resolves entrypoints, composes the selected module set, expands hard dependencies, validates reads/bindings, and hands off scoped work. Packages own their internal selection; pipelines own their procedures; the platform performs actual retrieval and tool execution.
The active root Catalog exposes exactly seven root packages, not every leaf. Zero, one, or multiple packages may contribute to different portions of one task. Selection is recomputed each turn and when task scope materially changes; it is not a permanent chat mode.
ROOT CATALOG KNOWS PACKAGE ENTRYPOINTS != ROOT CATALOG KNOWS EVERY LEAF
PACKAGE MEMBERSHIP != AUTOMATIC LOAD
PACKAGE SELECTION != ALL MODULES SELECTED
MODULE SELECTION != TOOL EXECUTION
PREVIOUS SELECTION != CURRENT SELECTION
The table summarizes the active repository contracts. Deployment and runtime evidence remain scoped separately below.
| Root package | Entrypoint / role | Contribution |
|---|---|---|
| TECHNICAL | HUB-TECH-ENTRY 1.2 / REASONING_PROFILE |
Mechanism reasoning, diagnosis with competing hypotheses, and bounded technical-change evaluation across subjects. |
| CONCEPT DEVELOPMENT | HUB-CON-ENTRY 1.8 / DOMAIN_CONTROL |
Concept consistency, causal development, bounded audit/analysis, and local selection of Creative, Brainstorm, Debate, Naming and task-applicable anti-drift controls. |
| AUTHORING | HUB-AUTH-ENTRY 1.0 / CAPABILITY |
Explicitly triggered, target-adapted prompt construction; produces a prompt and stops. |
| AIS | HUB-PIPE-AIS / TOOL_PIPELINE; single Project Source 1.3 |
External evidence verification, research, and provisional option discovery. |
| ImageT | HUB-IMAGET-ENTRY / TOOL_PIPELINE; controller 1.5 |
Image analysis and generation operations with input authority, prompt, approval, and visible-result checks. |
| SHOPPING | HUB-SHOP-ENTRY 1.2 / DOMAIN_CONTROL |
Product identity, purchase constraints, Offers and supported comparison; Hub selects AIS independently. |
| FOOD | HUB-FOOD-ENTRY 0.2 / DOMAIN_CONTROL |
Culinary analysis, cooking plans, adaptation, serving and scoped recommendations. |
TECHNICAL is selected for a reasoning need, not a technical topic alone. A diagnosis may follow symptom → competing hypotheses → mechanism chain → distinguishing test → supported cause → minimum sufficient change → behavioral verification. If evidence cannot distinguish causes, the cause stays unresolved. The module owns no technical subject domain, does not activate AIS, and does not turn a simple explanation into an architecture audit.
TECHNICAL TOPIC != AUTOMATIC TECHNICAL MODULE SELECTION
Concept Development is a growing package with its own local Catalog. The root Router selects the package; its entrypoint/Catalog select explicit operation leaves and task-applicable anti-drift controls under their current contracts.
| Concept leaf | Role | Premise boundary |
|---|---|---|
Creative 1.2 / HUB-CON-GEN-CREATIVE |
DOMAIN_CONTROL |
Develop inside the selected premise: imagery, action, rhythm, situations, relationships, and functional fictional detail. Fill gaps only within permitted creative latitude. |
Brainstorm 1.2 / HUB-CON-GEN-BRAINSTORM |
CAPABILITY |
Alter or branch the premise: reversal/relaxation, second-order consequences, feedback, and concrete bridges between incomplete premises. Develop alternatives in parallel; select a winner only when requested. |
If Creative needs a material premise change, it keeps that change conditional/proposed and returns the need through the concept package; it does not select Brainstorm or accept the new premise itself.
For illustration, a causal audit of an institution may need only the concept core; realizing daily life within it may add Creative; exploring alternatives to royal authority may add Brainstorm; realizing several premise variants may use both. These illustrate contributions, not literal routing rules.
The current Concept local Catalog admits five task-selected anti-drift controls: Anti-Prior, Modular Concept Architecture, Worldbuilding Internal Logic, Actor Reception and Narrative Resolution. Eligibility does not autoload them. Actor Reception Requires Modular Concept Architecture; the other controls add no hard edge. Historical A/B records retain their old revision scope.
AT is a prompt-enhancement capability: describe what you want in ordinary language, then explicitly choose the downstream target. It constructs a clearer, self-contained, target-adapted prompt from your intent, relevant current Hub context, and materially relevant evidence when available.
It is intended to help even if you know little or nothing about prompt engineering. You need not find a template online, learn prompt terminology, remember every context field, ask someone for “the right prompt,” or write a complicated structure yourself. AT preserves scope, constraints, uncertainty, evidence status, and target authority; it adds only the execution structure the task needs. It is scoped prompt construction, not a universal prompt optimizer.
For example, describe the research you want, then submit a message whose first non-empty user-authored line is:
/AUTHOR AIS
Put any additional request below that line. AT turns the intent and relevant context into a self-contained AIS prompt. The six supported commands are:
/AUTHOR AIS
/AUTHOR IMAGET
/AUTHOR FTH
/AUTHOR CODEX
/AUTHOR WORK
/AUTHOR GENERIC
The activation line must match exactly: case-sensitive, one ASCII space, no surrounding whitespace or extra text. Leading empty or space/tab-only lines may be skipped. Quoted/attached commands, code-block examples, later-line commands, and prior turns do not activate it. Reassess each turn; a clarification answer must repeat the trigger. Selection still requires the exact Library binding and full-read checks.
Hub discussion → /AUTHOR AIS → enhanced AIS prompt → user runs AIS separately
Fiction idea discussion → /AUTHOR GENERIC → prompt retaining the supplied fiction context
AUTHORING != EXECUTION
AT outputs the prompt and stops. It does not call AIS, FTH, ImageT, Codex, or Work. Authoring 1.0 is active in the repository. The CODEX routing smoke applies to the recorded earlier stack. See the module contract before use.
AIS is a Hub-selected evidence pipeline with three modes:
| Mode | Purpose |
|---|---|
VERIFY |
Check defined claims against external evidence. |
RESEARCH |
Collect, compare, and synthesize evidence within a scoped question. |
EXPLORE |
Discover externally grounded options that remain provisional. |
AIS EXPLORE != BRAINSTORM: AIS searches external sources; Brainstorm explores premise possibilities through reasoning. AIS returns evidence to Hub and owns neither the whole mission nor canon, design acceptance, or project authority. Greater research depth does not automatically launch Deep Research or Work. See the AIS pipeline.
ImageT handles image analysis, creation, editing, variation, correction, and retries—not just prompt writing. It controls prompt/source/reference authority and verifies the visible result. Selection alone generates nothing.
For generation operations, its high-level path is:
Hub selects ImageT → resolve current image controls → Open Gate
→ prompt construction / preservation → Prompt Gate → Ready Gate
→ user approves exact snapshot → one exact generation call
→ visible-result verification
Open Gate determines how much ImageT may transform the input: RAW permits scoped construction; PRESERVED_AUTHORED requires current explicit user authority to preserve the supplied semantic payload, with only minimal execution adaptation. Polished formatting or an AT-produced prompt alone does not establish preservation authority.
Ready Gate determines whether one exact generation snapshot is authorized to execute. Analysis alone generates nothing and bypasses Ready Gate. Changed inputs/sources invalidate the applicable approval; retries require a new snapshot and approval.
OPEN GATE != READY GATE
ONE APPROVAL = ONE EXACT TOOL CALL
See the ImageT controller and Ready Gate for exact contracts and verification limits.
REFINES specializes a matched responsibility while preserving uncovered duties. COMPLEMENTS lets independently applicable contributions coexist. REQUIRES is a hard technical dependency; only it automatically expands dependency closure.
REFINES != REQUIRES
COMPLEMENTS != REQUIRES
MENTION != DEPENDENCY
PACKAGE MEMBERSHIP != DEPENDENCY
Technical, Concept Development, Authoring, Shopping, ImageT and Food use Library-module bindings. AIS alone retains its Project-Source pipeline binding. Storage/binding establishes neither semantic hierarchy nor authority. Exact identities, full-read verification, deployment status, and scoped runtime records belong in the Central Hub documentation, Architecture, and Catalog.
User-reported build/test baseline: The current workshop controls and Central Hub architecture were developed and behaviorally tested primarily with GPT-5.6 Sol on ChatGPT Plus. This is the user's reported runtime qualification, not a compatibility result reproduced by repository checks. Other models have not yet been systematically tested; other plans/runtime conditions may work, but equivalent behavior and stability are not guaranteed. This does not establish incompatibility.
GPT-5.6 SOL + CHATGPT PLUS = CURRENT BUILD / TEST BASELINE (USER-REPORTED)
OTHER MODELS = NOT YET SYSTEMATICALLY TESTED
OTHER PLANS / DIFFERENT RUNTIME CONDITIONS = NOT GUARANTEED TO BEHAVIORALLY EQUIVALENT
REPOSITORY ACTIVE != PROJECT DEPLOYED
STRUCTURAL PASS != LIVE BEHAVIOR PASS
The model/plan baseline does not upgrade individual module or pipeline evidence. The active sources retain their specific deployment, retrieval, and live-test limits.
| I want to… | Go to |
|---|---|
| Continue or inspect one fiction project | Project Feather Core · The Kingdom · Fanfiction Engine |
| Take a local question into shared research, development, authoring, or tool workflows | Central Hub |
| Return Hub output to a fiction project | Use that project's own sources, controls, and review/admission rules |
| Explore a transferable fiction idea across independent projects | Fiction authoring model |
| Reuse LLM/worldbuilding controls | FTH Control Pack |
| Understand mission-oriented AI orchestration | Central Hub Architecture Pack |
| Read instruction/control designs and experiments | LLM controls |
| Read history and design analysis | Docs |
| Interpret the repository's sources and authority correctly | System context |
You can download either curated ZIP and use its included documentation without cloning the whole workshop. Each pack has its own release stream; neither version numbers the repository or proves an active personal deployment.
| Pack | What you can reuse | Get it |
|---|---|---|
| FTH Control Pack v0.1.0 | A control stack extracted from Project Feather Core for source authority, state, routing, and authoring boundaries. It contains no FTH lore. | GitHub Release · Source and adaptation guide |
| Central Hub Architecture Pack v0.1.0 | A reference architecture and optional starter for mission-oriented orchestration, scoped authority, and package composition. It is not a copy of the author's deployment; TECHNICAL is a replaceable worked example. | GitHub Release · Source and starter guide |
Start with the chosen pack's README, customize only what it requires, and test it in your own environment. The distribution landing page explains ZIP versus source use. Evidence belongs to each pack's release notes; LIVE VERIFIED: NOT_RUN for portability to another project.
The top-level folders separate the creative projects/systems from their supporting methods, instructions, workflows, and records:
repository/
├── README.md
├── SYSTEM_CONTEXT.md
├── AGENTS.md
├── LICENSE
├── worlds/
│ ├── The Kingdom/
│ └── Fanfiction Engine/
├── Project Feather Core/
├── authoring/
│ └── multi_paracosm_hub_model_updated.md
├── llm-controls/
│ ├── global-instructions/
│ │ ├── ChatGPT Plus+ Era/
│ │ └── ChatGPT Go-Free Era/
│ ├── project-instruction-designs/
│ │ └── Project CI/
│ └── worldbuilding-ci-kernel/
├── Central Hub/
├── Project CI/
├── distribution/
│ ├── README.md
│ ├── FTH-Control-Pack/
│ └── Central-Hub-Architecture-Pack/
├── codex-workflows/
├── docs/
│ ├── instruction-history/
│ ├── instruction-analysis/
│ ├── visuals/
│ └── assets/
│ └── brand/
│ └── logo-v1.svg
└── public-release-integration
- worlds/ keeps each project's internal sources, local controls, history, and tools together; the folder name does not classify Fanfiction Engine as a world. Moving a package does not change source status or authority.
- authoring/ contains the fiction authoring model: working material can move between independent fiction projects and Central Hub, and transferable results can be re-abstracted into fiction ideas before being explored elsewhere.
- llm-controls/ contains general instructions, project instruction designs, and reusable worldbuilding controls. Project-local controls also remain inside their world packages.
- Central Hub/ is the shared workspace for scoped capabilities and tool pipelines. Fiction projects may move work into and out of it without transferring local canon authority; it is not an additional creative system or a shared ontology.
- codex-workflows/ contains repository copies of Codex skills; moving them does not install or update the author's locally installed skills.
- docs/ contains instruction history, design analysis, explanatory visuals, and the workshop logo asset.
logo-v1is an asset revision, not a repository version. - distribution/ contains curated reusable projections with their own release streams; it does not own the active personal systems or their runtime bindings.
- Project CI/ is an existing workshop instruction-design area; its local controls and history retain their own scope.
The FTH instruction collection is managed with the Project Feather Core package. General instruction designs remain under llm-controls/; their separate responsibilities and historical baselines still apply. Directory placement does not select an active configuration.
The Kingdom package retains its declared reference-only status. Source archives retain their original contents and historical path references. Use current indexes for navigation; relocation does not reactivate archived sources.
The existing public-release-integration entry remains at the root as a Git commit reference (gitlink). Its role is unresolved here; it has not been moved, initialized, or treated as an ordinary content directory.
Important
LLM / agent readers: Read SYSTEM_CONTEXT.md before analyzing architecture, project relationships, or authoring workflow. Then follow the relevant project's sources and declared authority. Repository contents are a partial externalization of the author's systems; they do not establish the full internal world model or the currently active ChatGPT Project configuration.
The public packs are curated projections, not replacements for project-local sources. Keep repository checks, author-reported observations, and directly reproduced runtime behavior distinct. The sections below retain the workshop's deeper reading routes and interpretation boundaries.
Detailed reading routes and order
- SYSTEM_CONTEXT.md — authoring model, project boundaries, control responsibilities, and evidence rules.
- Central Hub — mission-oriented execution, routing, package composition, and tool-pipeline coordination.
- Project Feather Core — FTH sources, controls, current state, and project-specific authority.
- The Kingdom — its own world model and local rules.
- Fanfiction Engine — fanfiction/daydream simulation system; The Academy is its internal platform.
- ChatGPT Plus+ Era and ChatGPT Go-Free Era — Global Custom Instruction history for different character-budget eras.
- ChatGPT Project Global Native v9.1.1 — ACTIVE global baseline in the repository; design rationale and revision status. Project-specific version pins remain explicit; repository promotion does not deploy live instructions.
- Project CI and worldbuilding-ci-kernel — project-specific and reusable control designs.
- Instruction history and instruction analysis — design evolution, regressions, failure modes, and deployment notes.
Suggested reading order:
- Read this overview and SYSTEM_CONTEXT.md.
- Choose the creative system, Central Hub, or control question relevant to the task.
- Read that area's local documentation and authority rules before drawing conclusions.
- Consult history/failure analyses only when evaluating why a control or revision exists.
- Keep repository evidence, author-reported operation, structural validation, inference, and reproduced model behavior distinct.
- If the available material is insufficient, report the limit instead of completing missing lore or architecture from plausibility.
A search hit, filename, summary, directory placement, or repeated wording is not a substitute for reading the controlling source.
The sections below preserve the repository's provenance, interpretation boundaries, evidence model, and runtime limitations. They are intentionally more detailed than the overview above.
How work moves through the repository
Three connected flows explain the organization:
- Mission execution: a current task is scoped, relevant capabilities are selected, each contributor works within its authority, and the results return to the overall mission.
- Creative work: an idea or project-local question may develop inside one fiction project, move into Central Hub for shared capabilities, and return as working material for project-local interpretation and author review. A transferable local result may be re-abstracted into a new fiction idea and explored independently in another project; local implementation does not transfer with it.
- Control improvement: actual use reveals a failure; diagnosis leads to a candidate instruction or tooling change; author review, bounded implementation, checks, and live feedback inform whether the change is kept, revised, or rolled back.
These flows combine human actions, model instructions, and executable tools. They are not an automatic lore-publication pipeline. Git relocation does not update ChatGPT Project or Library contents, change live source-loading policy, or activate an instruction version.
Author and project premise
The author is not a professional software developer, data engineer, researcher, or member of a development team, and does not claim formal training in those fields.
This system developed through personal use, trial-and-error, observed LLM failures, repeated refinement, and practical attempts to keep several large fiction projects governable while using ChatGPT Projects and Codex.
Do not infer professional background, formal training, team structure, commercial intent, deadlines, external users, publication goals, or a formal engineering methodology from:
- repository size or complexity;
- frequent commits, branches, audits, or version history;
- routers, validators, rollback, provenance, tests, structured sources, or other engineering-like controls;
- software-like directory layouts or naming;
- data-like files or folder structures;
- dense infrastructure around an otherwise personal creative project.
These mechanisms exist because they became useful for personal LLM-assisted worldbuilding.
COMPLEX INFRASTRUCTURE != PROFESSIONAL DEVELOPER
HIGH COMMIT FREQUENCY != TEAM / DEADLINE / COMMERCIAL WORK
SOFTWARE-LIKE STRUCTURE != SOFTWARE PRODUCT
DATA-LIKE STRUCTURE != DATA PROJECT
ENGINEERING-LIKE CONTROL != FORMAL ENGINEERING BACKGROUND
REPOSITORY ORGANIZATION != AUTHORIAL ONTOLOGY
Evaluate the repository according to its stated purpose and observed failure modes, not by importing assumptions from professional software, data, enterprise, SaaS, or publication workflows.
What the workflow does
The repository name reflects its instruction-versioning history, but its current scope includes the creative systems, Central Hub, and the supporting control infrastructure. Global instructions may support general interaction as well as project work, while project-specific controls remain local to their respective projects.
A working session may continue fiction or simulation inside one project; take a local question into Central Hub for research, brainstorming, concept development, Authoring, AIS, ImageT, or other shared work; return that material to the fiction project for local review; externalize an accepted development into project sources; or diagnose a control failure exposed during real creative work. These are different uses of one personal workshop, not stages of one mandatory automated pipeline.
Different parts of this workflow have different roles:
| Repository term | Role in the creative workflow |
|---|---|
| Project sources | Record externalized world knowledge, decisions, and unresolved questions, with authority defined by the project. |
| Custom Instructions | Guide how the model interacts, reasons, and handles project material. |
| Router | Guide which sources the model should consult for a task. |
| Anti-drift controls | Address known ways the model can misread scope, authority, state, or causal relationships. |
| Source consolidation and validation tools | Organize source material and check declared structural requirements. |
| Regression checks | Check whether changes reintroduce known failures; what they prove depends on the kind of check performed. |
| Git history and review | Preserve revision history, inspect changes, isolate experiments, and support rollback. |
A pipeline diagram may describe author actions, model instructions, executable tooling, or feedback between them. It does not mean every step is automated. Running a repository check does not install instructions in ChatGPT, decide canon, or prove that a model will behave correctly.
How to distinguish this workflow from related repositories
Related repositories may combine prompt collections, worldbuilding templates, lore archives, writing tools, and evaluation workflows. The distinctions below explain this repository's scope; they are not a claim that no other project uses similar methods.
- Instructions are part of an ongoing authoring workflow. The repository records why a control was introduced, which failure it addresses, and what evidence supports it. A newer instruction file is a candidate to evaluate, not a universal upgrade.
- The creative projects/systems are independent. Project Feather Core, The Kingdom, and Fanfiction Engine do not share a mandatory architecture. Shared methods do not transfer canon, ontology, or source authority between them; Fanfiction Engine is not itself classified as a third fictional world.
- The author retains creative authority. Model output can support analysis and design, but plausible or repeated output does not become canon without the author's decision.
- The archive is a partial externalization. A world can contain author-established material that has not yet been written into this repository. Missing documentation limits what a reader or model can conclude; it does not prove that an element does not exist.
- Controls respond to observed failures. The infrastructure evolved through actual use, diagnosis, proposed repairs, review, and feedback. Its complexity should be assessed against the creative problem and recurring failure it addresses.
- Validation has explicit limits. File checks, source-consistency checks, and live model behavior are different kinds of evidence. A structural pass does not establish a runtime pass.
Read this as an author's working environment for maintaining fictional worlds with LLM assistance. Evaluate each tool and control by how it supports that work.
Interpretation boundaries
Do not assume:
- repository completeness;
- shared ontology across fiction projects;
- newer file = automatic authority;
- file placement = semantic ownership;
- basic lore examples = full canon;
- absent detail = absent from the author's internal world;
- plausible completion = accepted lore;
- structural validation = runtime validation;
- model output = authorial decision.
Evidence-state rules and the full authoring model are defined in SYSTEM_CONTEXT.md.
Evidence and version status
Keep these categories distinct:
- Structural validation: length, encoding, line endings, diff scope, declared metadata, or rule presence.
- Behavioral evidence: observed model output under stated test conditions.
- Author-reported observations: operational evidence that may not isolate every model, setting, memory, or conversation-history variable.
- Regression hypotheses: plausible failure paths that have not necessarily been reproduced.
A new instruction version is a testable behavioral hypothesis. A file's presence does not prove that it is deployed, approved, or authoritative. Proposed and unresolved material must retain its declared status.
Using instruction files
This repository does not automatically install, activate, inject, or apply instructions to ChatGPT, an account, or a device. Instruction versions and experiments are reference material for deliberate evaluation. Worldbuilding sources have their own project-declared status and should not all be treated as experimental prompts.
If you choose to reuse an instruction file:
- Read its scope, changelog, and known limitations.
- Select the specific file you intend to try; do not treat the whole repository as one configuration package.
- Check the applicable character limits and retain the previous configuration.
- Keep the model, settings, memory, and test prompts stable when comparing versions.
- Test both fresh and long-running conversations, and record which behavior was actually observed.
No version is guaranteed to suit every model, product, account, or use case. Copying or adapting instructions remains the reader's responsibility.
Independence and runtime limits
This is a personal project, not an official OpenAI project or a guarantee of ChatGPT behavior.
Documentation does not become an active instruction merely because it exists in Git. What is actually available and applied depends on the live configuration, instruction placement, source access, and conversation context. Repository inspection alone cannot establish that state.
The author decides canon. Tools and models help manage and examine the work; their output does not confer creative authority.