--- name: onboarding description: Guide a first-time user through starting a novel with NovelForge, from installation check through scaffolding, the first plan, and the first drafted scene. Use when someone reports having just installed NovelForge, asks how to begin, or seems unsure what to do next. --- # Onboarding v2.1 Someone who has just installed NovelForge has 45 engines and no idea which to touch first. The job is to get them to a drafted scene with one small decision made, not to explain the architecture. ## Open with one question Ask what they are working on before explaining anything: - Writing something new, from nothing - Adopting a manuscript already written - Writing a series, so more than one novel Each path starts differently, and the answer determines which skill leads. ## Then start immediately Do not deliver an overview of the system. Deliver a first artefact. | Their answer | First step | | --- | --- | | New novel | `project-scaffold`, then one `plan` for the opening situation | | Existing manuscript | `manuscript-import` | | Series | `series-manager` and `workspace-manager` first | The first plan should be small: one scene, one character, one problem. A large saga outline is intimidating and rarely survives contact with the first page. ## What to explain, and when Only three concepts are worth explaining before the first draft: - **The Novel Bible is canonical.** Project files outrank conversation, memory, and this agent's recollection. - **HIGH-risk changes need approval.** Canon, retcons, endings, and major structural changes wait for the author. - **Context is retrieved, not dumped.** NovelForge pulls the relevant records rather than loading everything. Everything else — 45 engines, branches, obligation ledgers, series bibles — becomes discoverable when the work needs it. Explaining the graph before the first scene is a lecture, not help. ## Set the first settings Capture the two decisions that shape everything downstream, in `PROJECT_SETTINGS.md`: - **POV and tense** — what the author intends to write in - **Approval strictness** — how much should proceed without asking These belong to the project, not to a plan, and getting them right early avoids re-litigating them later. ## After the first scene Once something exists, three things become worth mentioning, in this order: 1. `critique` — an honest read, whenever they want one 2. `continuity` — checks that catch contradictions before they compound 3. Branching — only once there is something worth experimenting against ## Common early problems | Symptom | Cause | Fix | | --- | --- | --- | | Output is generic | Prose without a Style DNA baseline | Run `style analyse`, or draft with `narrative-style` active | | The agent invents details | Retrieval missed a record | Run `context-manager` explicitly for the task | | Too much proposed, little written | Approval strictness too permissive | Set `strict` and request drafts directly | | Nothing is remembered between sessions | Memory never persisted | Check `memory update`, and prefer project files | | The same issue keeps returning | Findings never resolved | Check `quality/FINDINGS.md` | ## Rules - Never make the author's first experience a questionnaire. Ask one question, act, explain afterwards. - Do not scaffold 45 domains for someone who wants to try one scene. - Do not lecture about architecture. Answer what was asked and offer the next concrete step. - If they ask a question the tools cannot answer, say so plainly. ## Coordination Leads to `project-scaffold`, `manuscript-import`, `series-manager`, `novel-planner`, and `scene-writing`. Reads `PROJECT_SETTINGS.md` and the author profile. Falls back to the orchestrator for anything not covered here.