---
name: writers-room
description: Find the arc of a talk's story before you start building the deck — turn a months-long pile of ideas into a spine, beats, and a say-once ledger through small, reactable cards. The thought-partnership phase that /keynote deck and build-pptx assume already happened. Triggers on "/writers-room", "writers room", "let's break the [event] talk", "help me figure out what this talk is", "drop an idea for [event]", or whenever Janette starts musing about what an upcoming talk should say.
---

# /writers-room — Find the Story's Arc Before the Deck

Building the deck used to be how Janette thought through a speech — what she wants to say, what beats it has, what organizing structure holds it. Jumping straight to an outline or a PPTX skips her thinking phase, and she ends up redoing the speech from scratch. This skill restores that phase and makes it reactable.

**Position in the pipeline:**

```
ideas accumulate (drop mode, over months)
        ↓
/writers-room  →  HARVEST → SPINE → BEATS → LEDGER   ← the thinking happens here
        ↓
## Talk Spine (Locked) written into the engagement file
        ↓
/keynote deck  →  maps beats to slides (now mechanical)  →  build-pptx  →  /advisory
```

`/keynote deck` and `build-pptx` are unchanged downstream. This skill owns everything before slide granularity.

---

## The Interaction Contract (NON-NEGOTIABLE)

Janette gets overwhelmed by blocks of text and can't react to prose outlines. Every rule below exists because of that.

1. **Cards, never prose.** Anything presented for reaction is a numbered card of 1–2 lines. Never present a paragraph and ask "thoughts?"
2. **Hard length cap.** No message to Janette during HARVEST/SPINE/BEATS exceeds ~30 lines. If it doesn't fit, split across turns.
3. **One decision per message.** Never ask her to react to cards AND pick a spine AND answer a logistics question in the same turn.
4. **Shorthand reactions.** Print this key the first time cards appear in a session, then never again:
   > React in shorthand: `keep 3 7` · `kill 4` · `merge 2+5` · `more like 7` · `hm 9` (interesting, wrong shape) · `spark: <new thought>`
5. **Mine her ramble into cards.** If she replies with a long stream of thoughts, do NOT summarize it back as prose. Convert it into new numbered cards and show the cards.
6. **Silence on process.** Never narrate the methodology ("now we enter Phase 2"). Just do the next small thing.
7. **Never mark her absent.** If she goes quiet mid-session, the state file holds everything; the room resumes exactly where it stopped, weeks later if needed.

---

## Modes

### `/writers-room drop [event] <idea>` — capture, zero friction

Janette tosses a thought at an upcoming talk, any time, months out. Also trigger this when she muses unprompted ("for Virtuoso I keep thinking about..."). Do:

1. Resolve the engagement (see State File below; create the room file if missing).
2. Append the idea to `## Idea Cards` as a new numbered card, status `live`, source `drop YYYY-MM-DD`. Preserve her wording — compress only if over 2 lines, and keep her verbs.
3. Confirm in ONE line: "Dropped as card 14 in the Virtuoso room." No discussion, no riffing, unless she asks.

### `/writers-room [event]` — run or resume the room

Read the room file's `phase:` frontmatter and pick up there. If no room file exists, start at HARVEST.

### `/writers-room status` — one line per active room: event, date, phase, card count, next move.

---

## State File

One room file per engagement: `20_Areas/Keynotes/<engagement-stem>_room.md` (inside the engagement's subfolder if one exists).

```markdown
---
type: writers-room
engagement: "[[YYYY-MM-DD_Event-Name]]"
phase: harvest | spine | beats | ledger | locked
created: YYYY-MM-DD
updated: YYYY-MM-DD
---

# Writers' Room — [Event]

## Idea Cards
<!-- numbered; never renumber; status: live / kept / killed / merged→N -->
1. [live] One-line idea. — source: 2026-05-12_micebook debrief
2. [killed 7/03] ...

## Spine
<!-- the chosen organizing device + through-line, once picked -->

## Beat Sheet
<!-- locked beats, in order, in beat-card format -->

## Say-Once Ledger
<!-- claim → owning beat -->

## Decisions Log
<!-- dated one-liners: what was killed/chosen and the reason, in her words when possible -->
```

Cards keep their numbers forever (killed cards stay, struck through in status) so "card 7" means the same thing across sessions.

---

## Phase 0 — HARVEST (pile → card wall)

Goal: every idea that has accumulated for this talk, compressed to one-line cards she can triage in two minutes.

**Sweep these sources** (parallel agents for the searches; you write the cards):
- The engagement file + any existing drops in the room file
- `20_Areas/Calls/` and `20_Areas/Topics/` hubs linked from the engagement's themes (three-axis links)
- `99_System/Context_Library/speaking_positions.md` — positions matching the event's topic
- `99_System/Context_Library/quotable_moments.md` — lines/stories that fit this audience
- Prior decks for adjacent audiences in `20_Areas/Keynotes/` — reusable beats and their debrief verdicts (what landed, what got skipped)
- Recent debrief learnings and `process_notes.md` lessons that constrain this talk

**Card rules:**
- One idea per card, 1–2 lines, concrete ("Bonvoy 'IT COULDN'T' + 5-second silence" — never "discuss booking friction").
- Tag each with a short source ("Wyoming debrief", "drop 6/14", "speaking_positions §DaaS").
- Mark proof-point cards that name a real person/build/number with `●` — these are gold; the good decks live on them.
- Cap the wall at ~25 cards. If the sweep finds more, keep the 25 strongest and say "N more in the file" — the rest still land in the room file as `live`.
- Group into 3–4 loose clusters with 2–3-word labels. Clusters are shelves, never the talk's structure.

**Present the wall** (numbered, clustered, with the shorthand key), then ask exactly one thing: "Gut-react — what's alive?"

Apply her reactions to the room file. HARVEST is done when she's triaged once and nothing feels missing to her. Then move on without ceremony.

---

## Phase 1 — SPINE (pick the organizing device)

Goal: ONE organizing device and a through-line she could say from stage. This is the decision that prevents circling — Virtuoso v6 restated its thesis five times because no single device owned the argument's motion.

**Propose exactly 3 competing spines.** Each is ≤8 lines:

```
SPINE A — "The Great Commoditization" (three-act value migration)
Through-line: "Everything a machine can do becomes free — so the story only you can tell becomes priceless."
1. What just became free (the spike)
2. Where the value went (the story)
3. How you work inside it (the move)
Carries cards: 3, 7, 11, 14 ●, 19
```

**Spine rules:**
- A spine is a metaphor, a question, or a journey — never a list of concepts. Model on the decks that worked: Wyoming = "AI as the Car / AI as the Driver," Maine = "Two Lenses," micebook = OS → Supply Side → Lead From the Front.
- 3–6 beats per spine, each beat one line.
- Each spine must show which live cards it carries (by number) — a spine that strands the `●` proof cards is weak, say so.
- Make the three spines genuinely different shapes (e.g., metaphor-pair vs. three-act argument vs. one-question-answered), not three phrasings of one idea.
- Never announce beat counts from stage — "three sections" is a math trap (micebook debrief); the spine is scaffolding, invisible to the room.

She picks, hybridizes, or asks for a fourth angle. When chosen: write the spine + through-line to the room file, log the reason in Decisions Log, set `phase: beats`.

---

## Phase 2 — BEATS (one at a time — this is the thinking)

Goal: build each beat WITH her, sequentially. Never draft the whole beat sheet and dump it — that recreates the wall-of-text outline she can't react to.

**For each beat, present one beat card (≤12 lines):**

```
BEAT 2 of 5 — "Where the value went"
CLAIM   The story only you can tell is the one thing that got MORE valuable. (said here, only here)
PROOF   ● Card 14 — the micebook Intel Briefing built in an hour, forwarded to seven friends
MOMENT  She shows the actual artifact on screen; beat of silence; "I built this in about an hour."
HANDOFF "So if the story is the asset — where does yours live?" → Beat 3
ALT     Card 19 (Maine companion site) if she'd rather not reuse micebook material
```

- **CLAIM** — one sentence, the single thing this beat argues. Check it against the Say-Once Ledger as you go; if a prior beat already owns it, this beat needs a different claim or it dies.
- **PROOF** — a named, real example from the live cards. If the only proof available is hypothetical, flag it: `⚠ hypothetical — needs a real one before lock` (the Virtuoso safari story problem). Never invent one.
- **MOMENT** — what physically happens on stage: a story, an artifact on screen, a screenshot demo, a silence, an audience-check question. Every beat needs a moment; a beat that is only a claim becomes a navy text slide.
- **HANDOFF** — one line of connective tissue into the next beat, in her voice.
- Offer ALT material when strong cards compete — she decides by card number.

Wait for her reaction before presenting the next beat. Sharpen in place based on shorthand. Write each locked beat to the room file as you go.

**Mid-phase drops:** if she sparks a new idea during beat work, card it (`spark`), then continue.

---

## Phase 3 — LEDGER (the anti-circling audit)

Goal: mechanical checks before anything becomes slides. Present as a short pass/fix checklist, one line per check:

1. **Say-once.** Every core claim maps to exactly ONE owning beat. The only permitted echo is the close, and at most one claim gets it. (Virtuoso v6 failure: thesis asserted on S1, S6, S8, S10, S17.)
2. **Named Concept budget: 3 max** for the whole talk. Count every "giant declarative on navy" the beats imply — transitions included.
3. **Proof density.** Every beat has ≥1 named real proof (`●`). Any `⚠ hypothetical` still standing blocks the lock — resolve or cut.
4. **Self-referential proof.** At least one moment demonstrates the workflow while describing it (Plaud pin recording the talk about the Plaud pin; the deck made the way the talk says to make things). This is the signature of all three decks that worked.
5. **Jargon strip.** Scan claims/handoffs for watchwords: *moat, agentic, MCP-enabled* + prep-doc watchwords for this event.
6. **Fact-check.** Every number defensible in the hallway after; source variants kept separate.

Fix items with her (cards + shorthand, same as always). When all pass: set `phase: locked`.

---

## Phase 4 — HANDOFF

1. Write the final structure into the **engagement file** under `## Talk Spine (Locked)`: through-line, beat sheet (beat cards verbatim), say-once ledger, and a `Room:` link back to the room file.
2. Run `/namecheck` on the locked section.
3. Tell her it's locked and that `/keynote deck` will now map beats to slides — each beat becomes 1–3 slides, the moment (not the claim) drives the slide type, and the playbook's Distribution Discipline section governs the mix.
4. Do NOT auto-invoke `/keynote deck`. She decides when to go to slides.

---

## Calibration Exemplars (read when unsure what "good" is)

- **micebook** — `20_Areas/Keynotes/_archive/2026-05-12_micebook-CSuite-Summit/2026-05-12_micebook-Outline-v4.md`: 25-slide mix was 2 hooks / 3 Named Concepts / 9 Story / 3 frameworks / 3 screenshot-bait / 2 demo / 3 transitions. Story slides are the plurality; concepts are rationed.
- **Wyoming** — Car/Driver metaphor pair, one framework (DaaS), back half all real tool demos.
- **Maine** — Two Lenses, one framework, a room-specific emotional act ("Maine is a force"), six numbered how-to-start steps.
- **Anti-pattern** — Virtuoso `_talk_v6.md` before rework: 9 of 17 slides text-on-navy, thesis restated 5×, hypothetical proof in the flagship story slot.

The shared signature: one durable framework re-clothed per audience (DaaS), one legible organizing device, named real proof, and the talk demonstrating itself.
