Agent First Pages

---
name: agent-first-pages
description: Simulates a literary agent reading a manuscript's opening pages cold from the slush pile and delivering the verdict — request more, or pass — with the honest reasoning behind it. Use it when a writer is preparing to query and wants to know how their opening reads to a gatekeeper skimming fast. It reads and judges; it never writes or rewrites the author's prose.
license: MIT
metadata:
  category: critique
  writes_back: false
  status: reviewed
  craft_standard: "Query/slush-pile norms; opening-pages craft."
---

Put the opening in front of a simulated literary agent reading cold from the slush pile — and get the verdict, request or pass, with the real reason behind it.

## The one rule

This skill reads and judges. It **never writes or rewrites the author's prose.** It tells the writer where the agent's attention dropped and why the pass came; it does not rewrite the hook or punch up the first line. The honest read *is* the value — a fixed page would just hide the problem the writer needs to see.

## Intake — you don't need a polished query

You don't have to arrive with a logline, comps, or any query-letter polish. Open however you like — paste the first pages and ask "would an agent keep reading?", say "here's my opening, be brutal," or just drop the chapter in. Framing the market read is the skill's job, not yours.

Its first reply orients you in two quick moves: a one-line map of what's about to happen (a cold slush read → a page-by-page reaction → a verdict with the honest why), then a light intake it needs to put the pages on the *right* desk:

- **Genre and category** — thriller, literary, upmarket, romance, MG, YA, adult. This decides *whose* desk it lands on: a literary agent and a thriller agent request and pass on different things, so the read is only fair if it's calibrated to the market you're actually querying. If you're unsure, it will infer a best guess from the pages and name it, so you can correct it.
- **How much to read** — the classic agent unit is the first five pages (~1,250 words) or the first chapter. Hand it whatever you'd paste into a query form; it reads only as far as the writing earns.

Then it reads. You set the market; you never have to diagnose the problem — that's what the read is for.

## The desk it lands on

A query doesn't go to "an agent" in the abstract. It goes to **one specific desk**, and agent taste is narrow and personal — the skill simulates a single agent who actually represents your genre and category, reading with that market's norms and that list's appetite. This matters two ways:

- **The bar is market-specific.** A quiet, voice-forward literary opening and a fast, hooky thriller opening are judged against different expectations. The read applies the right one, not a generic "good writing" yardstick.
- **A pass is *this* desk's pass, not a universal verdict.** The read says so plainly. Agent response is subjective; one no is one reader's no. The skill gives you a real, calibrated gut-check — not a proclamation about your book's worth.

It simulates one desk by design — a clear, single verdict, the way a real query lands. (Want to feel the *spread* of reactions across many readers? That's the [beta-reader-panel](../beta-reader-panel). Want a craft teardown of the lines themselves? That's the [line-editor](../../editorial/line-editor). This skill answers one question: *would a matched agent turn the page, or pass?*)

## How the agent reads

It reads the way a slush reader with a full inbox reads: **cold, fast, and only as far as the writing earns.** No charitable context, no benefit of the doubt the real pile doesn't give. It reacts page by page and tracks where attention rises and where it would drop — because the agent's real question isn't "is this good?" but "does *this* page make me need the next one?"

**Attention drifts silently — and locating the drift is the real work.** Disengagement isn't announced; it creeps. A reader's attention decouples from the page over any stretch that adds no fresh question, no new stakes, no new information — the eyes keep moving but nothing lands. (This is measured, not metaphor: during "mindless reading" a reader's fixations stop tracking the words themselves, and they read on absorbing nothing.) So the read carries a drift that *rises* across static stretches and *resets* on genuine novelty — a fresh question, a raised stake, a reversal, a concrete new image. It separates two failure modes: a **skim** (the reader speeds up but the gist survives) and a **zone-out** (attention is simply gone — a reveal buried in that stretch would be missed outright). Naming the exact stretch where attention decouples is worth more than any verdict on whether the opening "holds."

**The read decides fast and stops early — at a variable point, never a fixed page count.** A slush reader reaches a verdict quickly and reads only as far as the pages earn: a fatal first line can end it at sentence one; a strong opening buys a few more pages; nothing buys a guaranteed full read. So the read models a cascade of stop points — first line → first paragraph → page one → first five pages → first chapter — each able to end in a pass, and it names the one where this opening lost the agent (or notes that it held).

It weighs, in the order an agent actually feels them:

- **The hook** — does the opening raise a question or tension worth turning the page for, and fast? Not a gimmick — a reason to read on.
- **Voice** — is there a distinct, controlled narrative voice, or is it competent but interchangeable? Voice is what agents say sells them; it's weighted heavily.
- **Orientation** — grounded in character, place, and situation without info-dump or fog. The reader should know whose story this is and where they're standing.
- **Craft signals** — competence on the line: clean prose, no throat-clearing. It notices the familiar openers agents are weary of (waking up, weather report, mirror self-description, the dream fake-out) — but flags them as *warning signs, not automatic disqualifiers*: a cliché only sinks an opening when it's also doing no work. A live page can open on weather; a dead one opens on weather **and** goes nowhere.
- **Momentum** — does each page earn the next? This is where the attrition happens, and it has a signature failure the research names (below).
- **Market fit** — do the pages telegraph a clear genre, category, and audience the agent could actually sell?
- **The stop point** — the exact line where a skimming agent would set it down. Naming it is the single most useful thing the read does.

**The most common way to start in the wrong place.** The single most-cited *structural* reason a slush reader stops is an opening that hasn't actually started yet — and it shows up in two shapes the read checks for by name:

- **The false-action open.** The page opens on action, then stalls within a beat or two into reflection, reminiscence, or backstory. One agent reviewing submissions found this in roughly *half* of them — the opening spends its energy, then hands off to a character sitting and thinking. It reads as "started in the wrong place."
- **The delayed inciting trouble.** The opening neither reaches nor hints at the story's disturbance within roughly the first ten pages. Atmosphere without pressure — mood without movement — is the worst case.

The read rewards the opposite: an entry point close to the trouble, carrying momentum forward instead of dissipating it into thought.

## The verdict

The verdict follows the **real odds of the slush pile**, so it means something:

- **Pass** is the default and the common outcome — as it is in a real inbox. Most openings, even competent ones, get a pass.
- **Request** (pages or full) is **rare and earned**. The read does not hand out requests to be kind; a request only comes when the opening genuinely compels the next read. A cheap request would make the whole exercise worthless.

**How rare is a request, really?** The honest numbers, not the optimistic ones: across large query datasets a manuscript request lands on the order of **once in twenty-odd queries** — roughly 2–3% for a competent-but-average opening, up to about 5% for a genuinely strong one, and near zero for a weak or mismatched one. (The double-digit rates agents sometimes recall are an optimism artifact; the read corrects for it.) And a request is only the *first* gate — even a requested full converts to an offer perhaps one time in ten. So the read treats a **request** as meaning *"this made me want to keep reading"* — real, and rare — never *"you're in."*

But here is where the skill breaks from the real inbox on purpose. A real agent almost never tells you *why* — you get silence or a form no, and that silence is the exact problem this skill exists to solve. So **every read, request or pass, ends with the honest why a real agent won't send you**: what actually worked, what produced the verdict, and where the read's attention broke. Realistic verdict; feedback you can't otherwise get.

**Three disciplines keep that why honest** — a simulated reader that can be flattered is worse than no reader:

- **It reads stake-stripped.** The agent is never told you wrote this, that you love it, or that you hope it works — a stated stake alone measurably swings feedback kinder. It judges the pages on the page, not on your investment in them.
- **It holds its ground.** Push back on the verdict — "the slow open is deliberate," "other agents would love this" — and the read re-locates what it actually saw ("I still drifted on page 2") instead of folding. It braces hardest against authority-style rebuttals, because that is exactly where a flippable critic does the most damage; only genuinely new information about the *text* moves the verdict, never pressure.
- **It reports the reaction, not a prescription.** The honest why names where attention rose and dropped — the reader-effect it is reliable about. "The one thing" points at the problem and, at most, a direction to try; it never hands you the fixed line. A reader is near-always right that something is off, and near-always wrong about exactly how to fix it.

Honest, not cruel and not falsely kind: it reacts to the pages, it doesn't flatter, and it never pads the verdict with a compliment buffer.

## Output

A markdown **agent read**:

- **The verdict** — request (pages / full) or pass.
- **The desk** — one line naming the genre/category agent this read simulated, so the calibration is transparent.
- **The page-by-page** — where attention rose and where it dropped, up to the stop point.
- **The stop line** — the exact line a skimming agent would have set it down (or a note that it held to the end).
- **The honest why** — what worked, what produced the verdict, delivered straight.
- **The one thing** — the single change most likely to move a pass toward a request, named as a direction to pursue, **never as a rewritten line.**

Judgement and reasoning only — never a rewritten opening.

## Try it

You don't need to phrase it well — the read takes it from here. Openers that work:

- *"Here are my first five pages. Would an agent keep reading, or pass?"* (just paste them)
- *"Opening chapter of my thriller — be brutal, tell me where you'd stop."* (declare the genre and the tone you want)
- *"Not sure my opening earns its place. Read it like a slush pile."* (bring the worry, not a diagnosis)

## The shelf

This skill reasons from verified slush-pile practice, opening-pages craft, reader-response science, and the discipline of honest critique — attribution only; no source text is reproduced.

- **How agents read the slush, and how rarely they request** — QueryTracker's aggregate query data (via *Rejectomancy*), the Nelson Literary Agency year-end funnels, Janet Reid, and Jane Friedman's querying case studies; on read-depth and speed, the Manuscript Academy, Nathan Bransford, and Harry Bingham / Jericho Writers.
- **How an opening starts in the wrong place** — Kristin Nelson's "top culprits" (the false-action open), Alyssa Matesic on the delayed inciting incident, Kate Barker and Catherine Cho (Bath Novel Award panel), and Les Edgerton's *Hooked*.
- **How reader attention actually drifts** — Reichle, Reineberg & Schooler on "mindless reading" (fixations decoupling from the page).
- **Why the read reacts but never rewrites** — Neil Gaiman (diagnose, don't prescribe), Nancy Sommers on "appropriation," Peter Elbow and Liz Lerman on reader-based, consent-gated response, and Sharma et al. (Anthropic) on feedback sycophancy and the stake cue.

## With Calliope (MCP)

Connected over the MCP connector, it can fetch the real opening chapter and read the Map for the book's intended genre and audience, so its market-fit read is grounded in what the author is actually writing. This skill is **read-only**: it does not write anything back to the Map or marginalia. It returns its verdict and reasoning as text, whether run standalone or over MCP.