Inkos

InkOS - Story Creation AI Agent

InkOS is a story creation and multilingual translation AI agent for long-form novels, short fiction, scripts, storyboards, interactive-film projects, fan works, continuations, covers, open-world / branching interactive play, and long-document localization. Prefer the Studio Chat / action-surface workflow for natural-language requests: the model proposes or invokes typed actions, InkOS executes them, and completion is derived from real tool results and files, not from prose claims.

Long-form writing still uses the chapter pipeline internally:

• Input governance: Architect / Planner / Composer preserve author intent, current focus, outline sections, and relevant truth files instead of injecting everything blindly.
• Writing: Writer produces prose with length governance, craft rules, and context selected for the current task.
• Settlement and review: Observer / Reflector update runtime state; Auditor checks continuity and quality; Reviser handles critical issues. The default write cycle keeps automatic repair conservative and leaves unresolved issues visible for human or agent follow-up.

Truth files are persisted as schema-validated JSON (story/state/*.json) with markdown projections for human readability. SQLite temporal memory database (story/memory.db) enables relevance-based retrieval on Node 22+. Persisted story memory is isolated to its project and book, excludes credentials and unrelated files, and is never reused across projects unless the user explicitly imports material. Users can inspect or delete the owning book/project through Studio or CLI.

v1.8.0 Mental Model

Treat InkOS as a pi-agent-centered production harness, not a bag of prompt shortcuts or parallel pipelines. The model interprets requests and emits typed actions; the host owns confirmation, deterministic tools, state, atomic persistence, and artifact truth. Standard AgentSkills / OpenClaw SKILL.md packages provide medium-specific craft and static references, but they never grant execution permissions.

• Natural-language requests should go through Studio Chat / TUI / inkos interact whenever possible.
• Production workers for long fiction, short fiction, scripts, storyboards, interactive film, Play, and translation use the same harness contract while binding different built-in Skills. Share the architecture, not long-form-specific prompts.
• Pipelines and atomic CLI commands are deterministic host capabilities. Do not create a second natural-language router around them.
• Do not infer success from assistant prose. A book, short, cover, or play step is complete only when the corresponding tool result and files exist.
• Use short_fiction_run only for a standalone short-fiction package.
• Use generate_cover only for cover generation/regeneration.
• Use play_start / play_step for Open World or Branching Interactive sessions.
• Use script / storyboard / interactive-film tools only for production artifacts that should be saved and exported, not for casual discussion.
• Use the translation project workflow for EPUB, text-based PDF, TXT, or Markdown localization. Keep source and target languages, glossary, review report, and export format explicit; do not replace it with an ad hoc one-turn translation when the user wants a complete deliverable.
• Use research_web only when the user explicitly asks for external facts, market references, era/profession details, or worldbuilding research. Research reports are reference material and do not automatically mutate canon or prose.
• Use long-form chapter tools only for existing long-form books.
• Use narrative forecasts when the author wants to compare possible long-form directions before writing. A forecast is non-canonical planning material: selecting a branch may write selected-branch-plan.md, but it must not be described as changing prose, outlines, or canonical state.
• A running production task does not prevent ordinary discussion, but do not start another conflicting book mutation until that task reaches a terminal state.
• Agent Skills provide professional guidance and static references. They do not grant new file, network, image, or writing permissions by themselves.
• Story memory, archived material, and Skill references share the local FTS5 / BM25 retrieval kernel. Source files remain authoritative; retrieved excerpts should retain source pointers.
• Bind reusable imported material to a book with explicit intended uses instead of repeatedly injecting every source file in full.
• Chapter prose, state, hooks, and run snapshots should commit through the safe chapter workspace / atomic file-set boundary. Never report completion from a partially persisted run.
• Context is governed: protected facts and current intent should not be silently compressed away; compressible history may be summarized when the context budget is tight.
• Studio Chat can receive user-uploaded text / Markdown / image attachments. Text attachments are injected into the LLM context; image attachments require a vision-capable model.
• External materials can be archived and retrieved later with evidence traces instead of relying on ad hoc pasted context.
• Prompt packs are user-tunable in Studio Project Settings. Project overrides are saved under prompt/<pack>/<prompt>.md; do not edit generated artifacts just to change system behavior.
• Long-form chapter revision from Chat passes the current user instruction into the reviser as a one-off brief. If the revision is not applied, inspect the returned gate metrics and remaining audit issues before claiming it was fixed.

v1.7 added multilingual translation/localization, narrative forecasting, background production, import, configurable review gates, recoverable locks, material archives, and standard Agent Skills. v1.8 unifies production around the pi-agent harness, adds medium-specific built-in Skills, shared FTS5 / BM25 retrieval, book-bound references, safe chapter workspaces, cross-format run snapshots, and TUI confirmation/model controls. Still surface unresolved review or execution issues plainly instead of claiming they were fixed.

When to Use InkOS

• English novel writing: Native English support with 10 genre profiles (LitRPG, Progression Fantasy, Isekai, etc.). Set --lang en
• Chinese web novel writing: 5 built-in Chinese genres (xuanhuan, xianxia, urban, horror, other)
• Fan fiction: Create fanfic from source material with 4 modes (canon, au, ooc, cp)
• Batch chapter generation: Generate multiple chapters with consistent quality
• Import & continue: Import existing chapters from a text file, reverse-engineer truth files, and continue writing
• Style imitation: Analyze and adopt writing styles from reference texts
• Spinoff writing: Write prequels/sequels/spinoffs while maintaining parent canon
• Standalone short fiction: Generate a complete short-fiction package with outline, draft, review artifacts, synopsis, selling points, and optional cover image
• Cover generation: Generate or regenerate only a cover prompt and cover image from a title, synopsis, or visual direction without rerunning story writing
• Interactive worlds: Start Open World or Branching Interactive sessions with world contracts, character agents, inventory/evidence/relationship state, guided choices, free actions, and optional image generation
• Interactive-film projects: Create playable branch graphs, variables/flags, relationship state, endings, node images, and exportable interactive project packages
• Scripts and storyboards: Convert ideas, outlines, or prose into script/storyboard deliverables while preserving user format choices
• Long-document translation: Translate EPUB, text-based PDF, TXT, or Markdown by chapter and semantic segment, maintain terminology, review source and target side by side, and export TXT/Markdown/EPUB