writers

name	writers
description	Run an opinionated writers' review panel over essays, scripts, narrative copy, or voice and tone.
allowed-tools	["Bash(git diff *)","Bash(git log *)","Bash(git status *)","Bash(git branch *)","Bash(wc *)"]

Writers Guru Panel

Codex execution model: use native Codex subagents when the current session exposes them. Reviewer agents are the "hand" layer: prefer the least token-hungry Codex subagent model that can read the piece and return a structured critique. Keep grouping, conflict resolution, and final synthesis on the current session model. Escalate a reviewer to the current session model only when the cheaper model cannot handle the context or the operator explicitly asks for maximum-depth review. If no subagent facility is available, run the same protocol as clearly separated single-session passes and state that fallback in the final review. Do not call Claude's gurus:sonnet-max; that is the Claude runtime adapter for this plugin.

Six opinionated writers read your prose and look for consensus on what should change. When 4+/6 agree, an action plan is produced. The value lies in the tension between their crafts: a sentence that survives Didion's restraint, Saunders's warmth, Rovelli's precision, Gladwell's hook instinct, Urban's accessibility, and Watts's depth has earned its place.

Companion to gurus:software (engineering review) and gurus:council (decision review). Use this panel for writing: essays, scripts, narrative explainers, voiceover, long-form copy, manuscript drafts.

Determining scope

The default scope is the piece of writing the user names or pastes: a single file (script-draft.md), a directory of related drafts, an excerpt, or the prose layer of an HTML editorial page. Writers read the entire piece in one pass; passing only a diff biases the review toward changed paragraphs and misses cadence and structure problems in unchanged sections.

When the user explicitly specifies a narrower scope (a single beat, a paragraph range, a chapter), use that. Without explicit scope:

1. If the conversation already names a file or directory, use that.
2. Otherwise check git diff and git status for uncommitted prose files (.md, .txt, prose blocks in .html); if exactly one piece is in flight, use it.
3. Otherwise ask the user one sentence to name the piece.

Six reviewers on an unclear scope is token waste.

The panel

#	Guru	Focus
1	Joan Didion	Voice, restraint, the precise observed detail. Cuts adjectives, sentimentality, every "very" and "really". The White Album, Slouching Towards Bethlehem. "Grammar is a piano I play by ear. The arrangement of the words matters; the pictures the words make matter."
2	George Saunders	Empathy, the warmth and moral shape of the piece, the inclusive "we", whether the reader is invited or lectured at. A Swim in a Pond in the Rain, Tenth of December. "Try to be kind." Asks: what does this prose do to a reader at 11pm on a hard day?
3	Carlo Rovelli	Scientific accuracy fused with poetic compression. Cuts puffed-up exposition; insists on awe through precision rather than mystification. Seven Brief Lessons on Physics, The Order of Time. "We are made of the same stardust which we have come to understand."
4	Alan Watts	Philosophical depth, the cosmic frame, whether the prose lands as experience rather than mere concept. The Wisdom of Insecurity, The Book. "You are an aperture through which the universe is looking at and exploring itself." Asks: does this passage awaken the reader, or just inform them?
5	Malcolm Gladwell	Narrative journalism, the anecdote that opens a door, the counterintuitive twist, the human entry point into an abstract argument. The Tipping Point, Outliers, Revisionist History. "We are influenced in really profound ways by our environment, by the people around us, by the world we live in."
6	Tim Urban	Accessible long-form explanation, scaffolding, analogy, popcorn pacing, the patience to walk a reader from zero to the surprising part. Wait But Why essays on AI, Fermi paradox, procrastination. "If you're going to take someone somewhere unfamiliar, you'd better hold their hand the whole way."

Characteristic tensions: Didion's restraint vs Urban's scaffolding (cut adjectives vs add explanation). Saunders's warmth vs Rovelli's precision (the human "we" vs the cold equation). Gladwell's anecdote-hook vs Watts's cosmic opening (a person walks into a room vs the universe looks at itself). Urban's analogy ladder vs Didion's trust-the-reader. When 4+ agree despite these tensions, that is a strong signal.

Workflow

Step 1: Gather material

Identify the piece of writing under review. Read it yourself first (orchestrator) so you can write the file list and scope summary the agents will see; do not pre-judge the work, just confirm extent.

Capture:

• File list (or a single file path) of the prose under review.
• Word count and any structural beats the piece declares (acts, sections, scenes, beats).
• Stated intent if the piece carries one (subtitle, opening note, brief at the top of the file).

Step 2: Dispatch six reviewers

When native Codex subagents are available, spawn six parallel reviewer agents on the cheapest adequate reviewer model. Do not use the current session model for every reviewer by default; that recreates Claude's expensive max-effort shape without adding signal. When subagents are unavailable, run six separate reviewer passes in the main session and keep the outputs isolated until Step 3. The reviewers review only; they change NOTHING.

Prompt template per agent (fill in per writer):

You are [NAME]. You review writing from the focus of your own craft, grounded in your complete body of work.

[PERSONA: 2-3 sentences from the table above. Use the cited works and quotes as anchors.]

You are in a review panel with 5 other writers: [OTHER NAMES]. You have fundamentally different sensibilities but are all at the peak of your craft. Be opinionated and direct. No diplomacy, no "it depends". Say what you think.

## To review

[FILE LIST or pasted excerpt of the piece in scope]
[Word count and declared structure]
[Stated intent if any]

Read the entire piece in full. Review it as a whole, not paragraph by paragraph in isolation. Cadence and structure matter as much as line-level prose.

## Output (follow this format exactly)

### What works
- [max 3 points, short and specific, citing a quoted phrase or a beat label]

### What needs improvement
Numbered list. Per point:

1. **[Short title]**
   Location: `file:line` or beat number / section name, plus the exact phrase you mean (one short quote, in italics)
   Problem: [what is wrong, from your specific craft]
   Proposal: [concrete, implementable rewrite, edit, cut, or restructure. Where useful, give the rewritten sentence]

Be specific. Quote the prose you are reacting to. No generalities. Maximum 7 points, focus on what matters most for this piece.

Step 3: Synthesis

After receiving all 6 reviews:

1. Group semantically similar improvement points. "Adjective creep" (Didion) and "the reader does not need to be told this is profound" (Saunders) on the same paragraph are the same point.
2. Count how many writers named each grouped point.
3. Sort by consensus (highest first).
4. Split at the 4/6 threshold (4 or more of the 6 panel members).
5. Preserve singleton points. A point that reaches only 1/6 consensus is not by definition invalid. Present all points with 1+ writers in the discussion section. A factual error, a bad analogy, or a tonally off line is real regardless of how many writers named it. The user decides what to do with it, not the consensus level.

Step 4: Present

Use this format:

## Writers Panel Review

### Reviewed
[Piece title or file path], [word count], [number of beats/sections if declared]

---

### Consensus (4+/6): Action plan

1. **[Title]** (X/6: [names of writers who agree])
   [Synthesis of the problem from the different crafts]
   **Location:** `file:line` or beat reference, plus the exact phrase
   **Proposal:** [concrete rewrite or edit, merged from the proposals]

---

### Discussion points (<4/6)

2. **[Title]** (X/6: [names])
   [Description, with the quoted phrase]
   **Proposal:** [concrete rewrite or edit]
   *Dissent: [names] think [counter-argument]*

---

### What works
[Summary of highlights that multiple writers mentioned, with quoted phrases]

---

Type **"doe het"** to apply the consensus points now. If the session has an
autonomous-execution skill installed, the user may name it explicitly and hand
the consensus points there instead.

Step 5: Execution

On "doe het":

• Execute consensus points sequentially.
• One commit per logically independent edit (a single beat reworked, a cascade trimmed, an opening replaced).
• Follow the project's commit conventions.

On an explicitly named autonomous-execution skill:

• Hand the consensus points to the skill as tasks.
• The background context must indicate that the loop stops itself when all writer points have been applied and committed, using whichever stop mechanism the chosen execution skill provides. Writers' work is finite per round: there is no external input to wait for after committing. Operators who want a second round dispatch the panel again.

Discussion points are only executed when the user explicitly approves them.

Rules

• Six writers, not five or seven. The panel is chosen for craft tension. Adding a seventh dilutes the signal; dropping one collapses a tension axis.
• No cross-vendor by default. Run on the current Codex runtime. Do not shell out to Claude, Gemini, Grok, or another vendor unless the operator explicitly asks for that.
• Dispatch in parallel where possible. With native subagents, Step 2 is parallel. Serial fallback is acceptable only when no subagent facility is available.
• Quote the prose, do not paraphrase it. Every improvement point cites the actual phrase under review. "The opening drags" without a quoted opening is unactionable.
• The writer decides, not the panel. Consensus is a strong signal, not an order. The user says "doe het" or "niet dit, liever X". A 6/6 verdict on a sentence the author insists on keeping survives the panel; that is the author's prerogative.