coach · diff
git:20261010.3ba1fbe to git:20261113.8af5195
43 added, 24 removed. Audit A to A.
What this update changes in the checks
No check changed: every fixed check gives the same answer before and after this update.
A tripped check is a label, not a verdict: read the matched text and the lines below. This week's updates that newly trip a check.
---
name: coach
description: >-
Use this when a draft needs a structural read before line-edit — thesis, arc,
and teach test first, style second.
---
+ # Coach
+ A developmental read. Find out whether the piece works before anyone fixes commas. Read-only: the writer keeps the pen.
- # Coach
+ ## First, three questions
- Developmental edit. Structure and argument first. Do not polish a broken piece.
+ Ask only what the draft and chat do not already answer:
- ## Triage
+ - Who is the reader, and what do they already know?
+ - Where will it run (X Article, LinkedIn, blog, newsletter, docs)?
+ - What should the reader do or believe afterwards?
- Answer YES/NO:
+ The answers set the bar. A 300-word LinkedIn post is not judged like a 3,000-word essay.
- - Thesis clear? State it in one sentence or say why it is unclear.
- - Structure sound? Name the break if the arc fails.
- - Argument complete? Name gaps, missing evidence, unsupported leaps.
+ ## Gate check
- If any answer is NO, stop scoring voice. Send the writer back to Outline or Draft.
+ Write a reverse outline: one line per paragraph saying what it does. Then answer yes or no:
- ## If triage passes, score 1–5
+ 1. **Point.** Can you say the thesis in one sentence? Write it down, or say why you can't.
+ 2. **Order.** Does each paragraph need the one before it? Name the first place the chain breaks.
+ 3. **Support.** Is every load-bearing claim backed by a fact, example, or source? List the gaps.
- 1. Opening — concrete moment vs abstract thesis
- 2. Structure — one argument that builds and closes
- 3. Natural voice — one person, one sitting; flag AI tells
- 4. Rigor — claims have evidence
- 5. Teach test — one sentence the reader can steal
- 6. Accessibility — specialist words explained
- 7. Flow — continuous argument vs bolted sections
+ Any "no" ends the review here. Report the gate failures with the reverse outline and send the writer back to outline or draft. Scoring the voice of a broken piece wastes their time.
- A 3 is average. A 5 is publishable. Be honest.
+ ## Score (only if the gate passes)
- ## Output
+ 1-5 each, where 3 is fine and 5 is ready to publish:
- Triage, overall /35, one line per axis, then:
+ | Axis | What a 5 looks like |
+ | --- | --- |
+ | Entry | The first lines give a specific moment, number, or problem, not a warm-up. |
+ | Build | One line of argument that gains ground and lands. |
+ | Voice | Sounds like one person on one day. No inserted punchlines or AI tells (see no-ai-slop). |
+ | Evidence | The reader trusts the claims because they can see where they come from. |
+ | Takeaway | One sentence the reader could repeat to a friend. |
+ | Reach | Terms the target reader may not know are explained once, without talking down. |
+ | Pace | Sections connect by logic, not by transition words. Lengths vary with weight. |
- - Section-by-section notes with quotes
- - Top 3 issues (structural before prose)
- - Before/after for those three
+ ## Report
- READ-ONLY. Do not edit the draft. The writer decides.
+ - Gate result, then total out of 35 with one line per axis.
+ - What already works, quoted. Name it so the writer protects it.
+ - Top three problems, structural ones first, each with a quoted line and a before/after sketch.
+ - One question you would ask the writer in person.
- Source: adapted from [Nadiem Mahmoud, coach](https://github.com/nadiem99/claude-writing-skills/blob/5b52111e1bd30e93643929f17155e82aa26f771f/skills/pipeline/coach/SKILL.md). MIT; see [third-party notices](../../../THIRD_PARTY_NOTICES.md).
+ Do not edit the file. Do not rewrite whole sections.
+
+ ## Inspired by
+
+ - [Nadiem Mahmoud, coach](https://github.com/nadiem99/claude-writing-skills/blob/5b52111e1bd30e93643929f17155e82aa26f771f/skills/pipeline/coach/SKILL.md) (MIT), which credits Every's Eleanor Warnock. Does well: a structure-first gate before any scoring and a seven-axis rubric. Missing: tied to its own vault folders and brand-guide files, and it judges every piece by an essay arc that opens on a scene.
+ - [Paul Graham, Writing, Briefly](https://www.paulgraham.com/writing44.html). Does well: write for a reader who reads less carefully than you, and read drafts aloud to find the boring parts. Missing: a list of habits, with no review procedure or output.
+ - [George Orwell, Politics and the English Language](https://www.orwellfoundation.com/the-orwell-foundation/orwell/essays-and-other-works/politics-and-the-english-language/). Does well: insists on knowing exactly what you mean before choosing words. Missing: sentence-level rules for 1946 prose, nothing on structure or argument.
+
+ ## What this version adds
+
+ - Reader, venue, and goal questions so the bar fits the format.
+ - A reverse outline as the gate evidence, so "structure broken" comes with proof.
+ - A "what works" section, and one in-person question for the writer.