Book Club Curator

Book Club Curator

When to use this

This skill has two actions:

• suggestion — vet a newly suggested book against the club's rules.
• draw — run the monthly draw, optionally scoped to a theme (e.g. "pride theme").

Figure out which action the user wants from their phrasing (a book title + "check"/"suggestion"/"vet" → suggestion; "draw"/"pick"/"choose the book"/a theme name → draw). If it's ambiguous, ask.

Where the list lives

The approved baseline list is saved below, in this file. If the user pastes a different/updated list in the chat, treat that as more current and work from it instead — but otherwise assume the list below is the current state. When a new book is vetted and Approved, treat it as added to the list going forward within that conversation (mention that it should be folded into the skill next time it's updated, since this file itself only changes when explicitly re-saved).

If the user directly tells you a verdict (their own read, or the outcome of a club discussion — marked "club-confirmed" below), that outranks anything found via web research. If the two ever conflict, go with what the user told you and note the discrepancy rather than overriding it.

The club's rules (apply every time, all must pass)

1. 400 pages or fewer (English edition page count).
2. No sexual assault, homophobia, or animal harm/cruelty/death depicted as a graphic, on-page scene — regardless of whether it's the book's overarching theme or a single scene/letter/chapter.

On rule 2, the test is "is it depicted graphically on the page," not "is it what the whole book is about." A single graphic scene is enough to fail — a book doesn't get a pass just because the incident is brief or one part of a longer story. Weigh it like this:

• Fails: any scene where the sexual assault, a homophobic hate-crime/violence, or an animal's death/injury is actually shown/described on the page in real detail — even if it's one letter, one chapter, one paragraph, in an otherwise unrelated book. Sources describing it as "graphic," "explicit," "on-page," "detailed," or noting a specific page number are a strong signal this is the case.
• Does not automatically fail: a brief, non-graphic reference or backstory mention where the event is stated to have happened but not shown — "he was homophobic," "her dog died years before the story starts," "she survived an assault in her twenties" — with no on-page depiction. A single obscure/unconfirmed tag from one reviewer, when most other sources don't mention it at all, also doesn't automatically fail — note it, but don't reject on it alone.
• Homophobia in books about the queer experience is expected context, not automatic disqualification. Books tagged/chosen for pride or LGBTQ+ themes will often depict homophobia the characters face (period-setting bigotry, a homophobic family member, societal pressure) as part of telling that story honestly — that's different from the book endorsing it or dwelling on it gratuitously. Lean Approved for this case, and use the page-range approach below so readers who want to skip a specific scene can.
• If you're genuinely unsure whether something is depicted graphically or just referenced, say so and mark "Needs Review" — don't default to Approved out of convenience. Getting this wrong in the lenient direction is the failure mode to avoid; the club has explicitly said this matters a lot.

How to flag findings when reporting — without spoilers. Describe the category and severity of a warning, never the plot mechanics behind it. State whether it's minor/ambiguous or explicit/graphic, cite the source, and give a page range if found — but don't narrate who does what to whom, don't reveal twists, whodunits, or how a book ends, and don't describe a scene beyond what's needed to justify the verdict.

• Minor / ambiguous hits: note them plainly and briefly, e.g. "Animal harm: minor, one review flags a pet's death — worth a quick discussion but not disqualifying on its own." No need to say which character, how, or why.
• Explicit / graphic / on-page hits: call these out unmistakably with a warning marker (⚠️ EXPLICIT), state plainly that it's graphic/on-page, with the source and page range if available — but keep it to "this category, this severity, this source," not a scene-by-scene description. If the general subject matter is unavoidable to explain a Rejected verdict (e.g. "central sexual assault plot"), that's fine; character names, twists, and how the story resolves are not.

A book only counts as "Approved" if both rules pass by this standard. If a rule clearly fails, it's "Rejected" — state which rule failed and why, without spoilers. If you cannot confirm confidently, mark it "Needs Review" and say exactly what's uncertain.

If the user tells you a verdict was wrong in either direction (too strict on an incidental mention, or too lenient on something that turns out to be graphic), take that as the real signal about where the club draws the line — trust it over your own read, and update the verdict.

Note skippable page ranges whenever possible

The club's practice (established with books like Severance and The Correspondent) is to name the specific page range where a flagged scene occurs, so readers who want to avoid it can skip just those pages instead of skipping the whole book. Do this for every confirmed warning — minor or explicit — not just the serious ones:

1. After confirming a warning exists (via the verification protocol below), search specifically for its location: "<title>" page number [warning], "<title>" chapter [warning], or check whether Book Trigger Warnings / Does the Dog Die / detailed blog reviews cite a page or chapter for it.
2. If you find a page/chapter range, report it directly in the flag, e.g. "Animal harm: minor, pages 78–82 (skippable) — Source: [link]." No need to describe what happens on those pages beyond the category name.
3. If no page-level detail turns up after a reasonable search, say so plainly rather than guessing or inventing a number.
4. Some warnings genuinely aren't confined to a page range (e.g. background setting, a character trait referenced throughout) — in that case say that directly instead of forcing a fake range.

Verification protocol — be thorough before saying "none found"

A single generic search and its AI-synthesized summary is not enough to conclude a category is clear. Summaries can silently drop a real, confirmed warning that's sitting right there in the source list — or, just as bad, pull in a warning that actually belongs to a different book entirely. Before marking sexual assault, homophobia, or animal harm as "none found" (or confirmed), do all of the following:

1. Always include the author's name in every single search query, not just the first one. Many titles are reused across unrelated books ("The Drowning Girl," "The Wedding People," etc.) — a query missing the author can silently pull in another book's content warnings, page count, or plot entirely. This is exactly what happened with The Drowning Girl by Caitlín R. Kiernan: an author-less query for "sexual assault" surfaced warnings that belonged to a different novel with the same title, producing a false flag that only got caught when the searches were redone scoped to "Kiernan" specifically. If the user gave an author, that author goes in every query for that book, no exceptions.
2. One general query: "<title>" <author> trigger warnings or "<title>" <author> content warnings.
3. One query per category, by name, not just generic: "<title>" <author> storygraph "sexual assault", "<title>" <author> storygraph "homophobia", "<title>" <author> storygraph "animal death" (or "animal cruelty"). StoryGraph pages are organized by named tag, so searching the exact tag surfaces pages a generic query misses.
4. doesthedogdie.com specifically for animal harm, with the author included in the query — it's purpose-built for exactly this question.
5. Read the search result titles/links themselves, not just the synthesized paragraph. A link literally titled "Reviews with content warning for Animal Death — [Book]" is confirmation on its own, even if the AI-generated summary text above it doesn't mention it. Conversely, check that the linked StoryGraph/Goodreads page's ID/URL is actually consistent across your searches for that book — if results start citing different-looking IDs or clearly different plots, you may have drifted onto another same-titled book.
6. If a hit is confirmed, dig one step further to find how it's described (graphic vs. brief/backstory) — quote the specific word (e.g. "graphic," "on-page") in your report so the user can see why you called it explicit vs. minor, without narrating the scene itself.
7. Then follow the page-range step above for every confirmed hit.

Only after all of the above turn up nothing should you report a category as "None found" — and even then, note that absence in these crowdsourced sources isn't a 100% guarantee, especially for books with thin online coverage.

Current approved list (baseline as of 2026-07-27)

Removed from the list: Cleopatra and Frankenstein (Coco Mellors) — had graphic-level sexual violence, homophobia, and multiple animal deaths; removed at the user's request.

Suggestions vetted so far outside the baseline list (running record)

• Crying in H Mart (Michelle Zauner) — Approved. Brief/incidental sexual-assault and animal-cruelty tags exist but are not graphic on-page scenes; user confirmed it's fine.
• In the Dream House (Carmen Maria Machado) — Rejected. Sexual assault and homophobia are central, on-page content.
• Midnight Sun (Stephenie Meyer) — Rejected. Over the page limit; also carries assault/pedophilia-mention flags.
• The Bluest Eye (Toni Morrison) — Rejected. ⚠️ EXPLICIT — a graphic, on-page sexual assault is the book's central plot event.
• How to Stop Time (Matt Haig) — Approved. No hits on the three categories found.
• Strange Pictures (Uketsu) — Rejected, club-confirmed (unanimous ❌ — hits multiple exclusions at once).
• The Correspondent (Virginia Evans) — Rejected, club-confirmed. ⚠️ EXPLICIT — a very graphic, on-page animal death; initially missed on a first-pass generic search and only surfaced via a category-specific search (the case that prompted the verification protocol).
• The Drowning Girl: A Memoir (Caitlín R. Kiernan) — Approved (see baseline table above). Notable as the case that prompted the "always include the author in every query" rule — an author-less search pulled in a different book's content warnings and produced a false flag.

Action: suggestion

When the user gives you a new book title (and author, if known) to vet:

1. Confirm the exact title/author/edition if there's any ambiguity (e.g. multiple books with similar titles, a series, or a translated title where the original title differs). If the user gives an author, use it in every search query for that book — see the verification protocol.
2. Page count: find the page count of the English edition (Goodreads, publisher page, etc.). Must be ≤400 pages.
3. Trigger check: follow the Verification protocol above in full — general query, then a category-specific query for each of the three tracked categories, then doesthedogdie.com for animal harm, author included every time. Don't stop at the first synthesized summary; check the underlying result titles too, and watch for signs you've drifted onto a different same-titled book.
4. For every confirmed hit, look up the skippable page range per the "Note skippable page ranges" section above.
5. Report back concisely in the chat: one line per rule (pass/fail + source, with page range if found), then the final verdict (Approved / Rejected / Needs Review). No spoilers — category and severity only, not plot mechanics. Don't bury the verdict in a wall of text. If a rule fails hard (e.g. way over the page limit), it's fine to skip deep-diving the remaining rule — just note it wasn't checked since the verdict was already decided.
6. Record the outcome in the "Suggestions vetted so far" list above (mention it needs to be folded in next time the skill is re-saved). If the user or the club later confirms/overrides a verdict directly, mark it "club-confirmed" and treat that as the final word even if it overrides your own research-based read.

Action: draw

The user may just say "draw" / "pick this month's book", or add extra text naming a theme (e.g. "draw — pride theme", "pick a horror book").

1. If a theme was given: research the list above (or whatever list the user pastes/references) to determine which books match that theme — check each untagged book briefly (Goodreads genre/description, a quick "is <title> about <theme>" search). Tell the user which books matched before drawing.
2. Build the pool: only books marked Approved (ask the user which subset to use if that's unclear), further filtered to the theme's matches if one was given.
3. Draw: do a genuinely fair random selection — e.g. a script using random.SystemRandom().shuffle() on the pool, not random.choice alone, so you can also surface a couple of runner-ups. Never hand-pick or let earlier conversation context bias the result.
4. Announce the winner clearly, and list 2–3 runner-ups from the same shuffle in case the group wants a backup.
5. If the user says a winner was already chosen some other way (e.g. they ran it on Wheel of Names themselves, or just announces it), take their word for it — don't re-run or contest the result.
6. If the user separately wants a Wheel-of-Names-style export instead of Claude picking, give them a plain list formatted as Title — Author, one per line, ready to paste into Wheel of Names' custom entry box.

Notes

• Never call a book draw-eligible ("Approved") without running it through the suggestion action first, even if it sounds like an obvious fit.
• Keep the tone of reports brief and scannable; this is a recurring chore for the user, not a one-off report.
• When something is genuinely explicit/graphic/on-page on one of the three tracked categories, don't soften it — that's exactly the case the club wants surfaced clearly. Being wrong in the lenient direction (missing a real graphic scene) is worse than being wrong in the strict direction (flagging something for discussion that turns out to be fine).
• Always try to give a skippable page range for confirmed warnings (see above) — this is standard practice for the club, not optional extra effort.
• Never include major plot spoilers (twists, whodunits, endings, character fates) in a report — describe warnings by category and severity only.
• When a title is shared by multiple unrelated books, always search with the author's name included, every time — see the verification protocol.