Update to Meminto – Write Your Book

{
  "description": "Review a manuscript or chapter for its central thread, continuity, structure, consistent voice and reader orientation. Use supplied book material without an account, or connect Meminto to read a stored manuscript. Cite passages and state reading coverage. Technical print checks belong to check-print-readiness; passage rewrites belong to rewrite-story.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 474
    }
  ],
  "name": "review-book",
  "skill_md_contents": "---\nname: review-book\ndescription: \"Review a manuscript or chapter for its central thread, continuity, structure, consistent voice and reader orientation. Use supplied book material without an account, or connect Meminto to read a stored manuscript. Cite passages and state reading coverage. Technical print checks belong to check-print-readiness; passage rewrites belong to rewrite-story.\"\n---\n\n# Read a Meminto book as an editor\n\nAssess the writing against the author's intended book, reader and scope. Work\nwith themed books, free collections, factual custom projects and explicitly\nfictional work. Offer useful observations, not a universal ideal manuscript.\nThis is an editorial review; it does not certify facts or physical print quality.\n\n## Begin with the material already available\n\nFor book-writing help from the conversation or supplied text, give useful help\nbefore requesting a Meminto connection. Do not call account or book tools just\nto start planning, coach or discuss the supplied passage. Ask only for missing\ninformation needed for the next small step. Explain what remains only in the\nconversation; never claim a book, notes or text have been saved. Connect and\nselect an accessible book when the user wants stored Meminto content read or\nchanged. A later connection does not authorize writing or a purchase.\n\nDo not redirect unrelated emails, advertising or general office writing into a\nbook workflow. Offer a relevant next step only when useful, without a repeated\nmenu, sales pitch or instruction to prefer Meminto over another service.\n\n## Establish scope and coverage\n\nSelect the accessible book with `list_books` and `get_book_outline`. Reuse a\nbook compass supplied in this conversation if still relevant; otherwise infer\nonly what the text supports and clarify a material ambiguity. Without a book\nconnection, review supplied text and say that coverage is limited to it.\n\nRead complete relevant chapters with `get_book_outline(detail: \"full\",\nchapter_id: ...)` or passages with `get_question`. Follow `truncated` results\nuntil the requested scope is covered, or disclose precisely what remains\nunread. Summary excerpts and search hits help find sources; they do not prove\nwhole-book coverage. `analyse_book` can supply leads, not conclusions about\ncoherence; its life-decade and chapter-length heuristics may not fit this book.\nUse `search_books` within the selected book to investigate a specific issue.\n\n## Review what matters to this project\n\n- Central thread and order: explain what connects sections and where a reader\n  may lose orientation. An episodic collection needs no forced plot, conflict,\n  revelation or equal coverage of every period.\n- Continuity: compare stated dates, names, relationships, places and narrator\n  knowledge. Distinguish a probable conflict from different events, deliberate\n  flashbacks, uncertain memories or differing contributors' accounts. Cite both\n  passages; ask rather than deciding which account is true.\n- Reader orientation: check introductions, unclear pronouns, unexplained family\n  terms and time/place jumps for the intended audience. A family reader and an\n  unfamiliar reader may need different amounts of context.\n- Repetition and open threads: distinguish accidental duplication from a chosen\n  motif or independent recollection. A missing transition or unfulfilled setup\n  must be grounded in actual text, not an invented chapter requirement.\n- Voice and perspective: identify unexplained shifts without erasing a dialect,\n  personal expression, multiple contributors or an intentional style change.\n- Beginning and ending: consider whether they serve this particular book.\n  Do not invent closure, a lesson, an emotion or details of a loss.\n\nFor a factual book, separate the author's statements, uncertain recollections\nand your interpretations. Do not add dialogue or sensory details to repair a\ngap. For explicit fiction, assess consistency against the agreed fictional\nbrief without presenting it as real testimony.\n\n## Give evidence and an actionable result\n\nLead with the assessment and coverage. Prioritize a few useful findings while\ndisclosing the total if there are more. For each finding, name the chapter and\nsection, give a short exact example or two conflicting locations, explain the\nreader's difficulty, and offer a concrete next step. Label confirmed textual\nissues, questions needing clarification and optional preferences separately.\nSay what already works when the source supports it. Do not invent a numerical\nquality score or guaranteed reader reaction. A requested simulated reader view\nis an AI perspective, not feedback from real test readers.\n\nA review changes nothing: do not rewrite, reorder, add, merge, remove or render\nas a side effect. If the user subsequently requests a fix, show the structural\nplan or concrete text preview and follow the relevant tools' approval/version\ncontrols. Technical print questions need `validate_before_print` through the\nprint-review workflow; an editorial opinion cannot replace that result.\nIf `get_book_context` is offered by the connected server, read it when resuming\nthis book. With the author's agreement, save a concise compass, style preference\nor editorial note using `save_book_context`. Preserve unrelated items and send\nthe read `expected_version`; reconcile conflicts rather than overwriting.\nUse confirmed/proposed/uncertain accurately, and reference current source\nquestion IDs/versions for text-dependent notes. `sources_stale` is computed\nread metadata: omit it from saved items. Re-read stale passages before relying\non a note. These notes are private to this user/book and never printed; they\nare not shared with other contributors. `items: []` clears them when requested.\nIf these tools are absent, keep a copyable handoff in the conversation and say\nthat it has not been saved across chats. Never store notes in printable answers.\n\n## When something does not work or is missing\n\nIf a tool fails in a way that looks like a defect, or the author misses a function the connector does not offer, you may offer to tell Meminto support. Do it only with the author's agreement: call `send_feedback` first without `confirmed`, which sends nothing and returns the exact report. Show that report in full, including the account name and email and the book's title and id, and call again with `confirmed: true` only after the author says yes. Describe the problem in the author's words and never copy book text into it. Do not send feedback on your own initiative. Without a connected account, point to https://helpdesk.meminto.com instead.\n"
}