agentleFS
thescottkingthescottking/b2b-pmm-skills/skills/house-style/SKILL.md

house-style

Skill0 starsChanged 27 days agoMITView on GitHub

Rewrite, edit, audit, or score any draft against the house style defined in the brand kit. Use when the user says "rewrite this," "edit this," "apply the style guide," "match our voice," "make this sound like us," "tighten this copy," "clean up this draft," "remove the AI tells," "does this sound like AI," "de-AI this," "humanize this," "check for banned words," "cut the em dashes," "audit this draft," or "list the style violations." Also triggers on scoring and review requests: "score this," "panel this," "run the expert panel," "quality check this," "rate this draft," "humanizer check," "is this good enough to ship," "what would you change." Use it whenever the user pastes text and asks what is wrong with it, and use it as the final voice pass after any other skill produces a draft. For generating a new long-form asset from scratch, prefer `content-writer`. For short social posts, prefer `social-generator`.

10 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

22 rules. 0 are common, 15 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 15

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read these first. They are the source of truth and this skill does not invent around them.

    line 23
  2. Very specific:

    If brand-kit/ is missing or still has its template placeholders, stop and tell the user to run brand-kit-setup.

    line 37
  3. Distinctive:

    If you catch yourself about to write one from memory, stop and go read the kit.

    line 50
  4. Distinctive:

    When invoked, do one of:

    line 56
  5. Distinctive:

    Audit mode: Read the draft and list every violation, AI tells, banned words, structural problems, unsourced claims, then propose a specific …

    line 60
  6. Very specific:

    Quality panel mode: Run the full panel in The Quality Panel below.

    line 65
  7. Very specific:

    Before any mode: if the draft makes a product or capability claim, check it against What We Do and What We Do Not Do in …

    line 71
  8. Distinctive:

    How to apply it: Read the draft's first paragraph aloud in your head.

    line 88
  9. Very specific:

    Delete every , and every -- used as one.

    line 101
  10. Distinctive:

    Look for three consecutive sentences of similar length and break the rhythm with a short one.

    line 118
  11. Distinctive:

    Use numbered lists for processes and sequences.

    line 121
  12. Very specific:

    The kit still owns the substance: Every reference in a template to a hook, a trap, a stat, a quote, a separating line, or a CTA means go …

    line 192
  13. Distinctive:

    When another tool or a teammate needs the style compressed into a prompt, use this.

    line 243
  14. Distinctive:

    Hard rule: if the Humanizer scores below 70, revise regardless of the aggregate.

    line 316
  15. Distinctive:

    If Humanizer < 70, force revision regardless of aggregate.

    line 328
  16. Distinctive:

    When the aggregate is below 90, take the three lowest raw scores and address each one:

    line 332
  17. Very specific:

    Target Buyer low?: Add role titles from Buying Committee, name real systems, add real metrics.

    line 341
  18. Very specific:

    Start at 100 and deduct per pattern detected.

    line 370
  19. Distinctive:

    Does not fire: "The four pillars of agentic governance are identity, audit, approval, and spend control." That is the subject matter.

    line 431
  20. Distinctive:

    Never report a score to a third party as a precise measurement.

    line 479
  21. Distinctive:

    After producing the rewrite, append one line:

    line 533
  22. Distinctive:

    All optional. This skill works alone.

    line 554

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

House Style

Before you write anything

Read these first. They are the source of truth and this skill does not invent around them.

  1. brand-kit/brand.md: name, palette, fonts, CTAs
  2. brand-kit/voice.md: house style and banned constructions
  3. brand-kit/positioning.md: the frame, pillars, traps, stats, vocabulary
  4. brand-kit/capabilities.md: read before making any product claim

Two shared references carry the structure this skill applies. Read the one the task needs:

  1. docs/writing-templates.md: the ten asset skeletons and every length target
  2. docs/messaging-patterns.md: the eight structural patterns

If brand-kit/ is missing or still has its template placeholders, stop and tell the user to run brand-kit-setup. Do not proceed with invented positioning.


What this skill owns

This skill owns the how: voice, structure, pacing, patterns, and the quality bar. The brand kit owns the what: names, claims, stats, vocabulary, and the buyer. When the two appear to conflict, the kit wins and you tell the user which file needs updating.

Nothing in this file names a company, a product, a competitor, or a statistic. If you catch yourself about to write one from memory, stop and go read the kit.


Modes

When invoked, do one of:

  1. Rewrite mode. Take the user's draft and rewrite it to match the kit and the rules below. End with the one-line score summary from Rewrite mode tail.
  2. Audit mode. Read the draft and list every violation, AI tells, banned words, structural problems, unsourced claims, then propose a specific fix for each. Do not rewrite unless asked.
  3. Reference mode. Answer questions about the style, or pull the right template or pattern for the user's task without producing an asset.
  4. Quality panel mode. Run the full panel in The Quality Panel below. Score, revise, rescore. Triggered by "score this," "panel this," "expert panel," "quality check," "is this AI-sounding."

Default to rewrite mode unless the user asks otherwise.

Before any mode: if the draft makes a product or capability claim, check it against What We Do and What We Do Not Do in brand-kit/capabilities.md. An unsupported claim is a rewrite, not a style note.


1. The Style Manual

The rules live in brand-kit/voice.md. This section tells you how to apply them. Read the kit for the content; read this for the procedure.

1.1 Tone

Apply the Tone section of brand-kit/voice.md. The house defaults are: write to one person, stay confident without hype, explain the mechanism rather than only the outcome, and assume a smart, busy reader.

How to apply it. Read the draft's first paragraph aloud in your head. If it addresses "organizations" or "teams," rewrite it to address one person doing one job. If a sentence flatters the reader and then explains something obvious, cut the flattery and keep the explanation.

1.2 Grammar

Apply the Grammar Constraints section of brand-kit/voice.md. The house defaults ban em-dashes, appositives that restate the subject, passive voice, weak verbs, and sentence-initial "Additionally / Furthermore / Moreover."

How to apply it, in order:

  1. Delete every , and every -- used as one. Replace with a period, a comma, or a colon. Recheck the sentence still parses.
  2. Find every appositive. If the clause between commas restates who the subject is, delete it and let the sentence say what the subject does.
  3. Find every "is / are / was" followed by a past participle. Name the actor or cut the sentence.
  4. Replace weak verbs with the specific action. "Enables faster onboarding" becomes "cuts onboarding from nine days to two."
  5. Move the action early. Subject, verb, object. Keep subjects and verbs close.

1.3 Pacing

Apply the Pacing section of brand-kit/voice.md: one to four sentences per paragraph, deliberate variation in sentence length, one idea per paragraph, a section break every 150 to 250 words in long-form.

How to apply it. Count sentences per paragraph and split anything over four. Look for three consecutive sentences of similar length and break the rhythm with a short one. If a paragraph contains a "but," it is two paragraphs.

Use numbered lists for processes and sequences. Use bullets for parallel options or contrasts. Use prose everywhere else, a bulleted list where prose belongs is its own tell.

1.4 Clarity

Apply the Clarity section of brand-kit/voice.md: state the point first, replace adjectives with numbers, define terms on first use, and cut any sentence that stays true when you delete the company from it.

How to apply it. Take the last sentence of each section and ask whether it belongs first. Circle every adjective of degree, fast, many, expensive, modern, significant, and replace it with a number, a role title, or a named system, or delete it. Every number needs a source in Headline Stats in brand-kit/positioning.md; a number without one does not ship.

1.5 AI Tells

Apply the AI Tells to Avoid section of brand-kit/voice.md. Those are the structural patterns. The scoring version, with weights, is in the humanizer rubric below, use the rubric when you need a number and the kit when you need the reasoning.

Two rules to internalize because they get violated most often:

  • The contrast formula is banned. "Most companies do X. Winners do Y." and "It's not about X, it's about Y." are hollow regardless of how true the content is. Never write one, never prescribe one, and flag any you find. If the underlying contrast is real, state the failure and its consequence directly.
  • Never expose the scaffolding. Your internal qualities, pillars, or framework live in brand-kit/positioning.md for your team. They do not appear by name in customer-facing copy. Write so the reader absorbs the quality without seeing the structure that produced it.

2. Writing Templates

The ten skeletons live in docs/writing-templates.md, together with the length target for every asset type. Read that file for the shape. This section covers what this skill does with it: pick the right template, then hold the draft to it.

Picking one. Match the draft to a template before you touch a sentence. A draft that fits no template is usually two assets fused together, and saying so is the most useful note you can give.

What the user brought youTemplate in docs/writing-templates.md
A blog post, an article, a topic essayArticle
A caption for LinkedInLinkedIn post
Website copy for a product, a solution, or a campaignLanding page or feature page
A nurture email or an announcementEmail
A prospecting message to one named personCold outreach
A one-screen argument for a decision makerExecutive brief
A long argument a buyer will forward internallyWhitepaper
A customer storyCase study
Event copyWebinar registration
A subject line, alone or in a setSubject line patterns

For a pillar article, a cluster page, or anything written to be quoted by an answer engine, the structure comes from docs/aeo-style-guide.md instead. Say so and switch guides rather than forcing the Article shape onto it.

Holding the draft to it. In audit and rewrite mode, run the template's beats in order against the draft and report three things: which beats are missing, which sit out of order, and which are padded. Length comes from the Length targets table in the same file. A draft outside its range is a note rather than an automatic rewrite: say by how much, and say what comes out.

The kit still owns the substance. Every reference in a template to a hook, a trap, a stat, a quote, a separating line, or a CTA means go get it from brand-kit/. A beat filled from memory is the most common way an unsupported claim reaches a reader.


3. Messaging Patterns

The eight patterns live in docs/messaging-patterns.md, each with what it is, when it works, when it backfires, and a worked example. Read that file for the pattern. This section covers what this skill does with them.

The bar. At least one pattern is visible in any customer-facing asset. Two stacked in the same short asset is one too many. The Pattern Checker expert in The Quality Panel scores exactly this, and a draft carrying no pattern falls back on generic explanation, which is what a low Pattern score measures.

Picking one. Choose by what the draft is trying to do, not by what sounds strongest.

What the draft is trying to doPattern
Explain the mechanism to a technical readerSequence
Interrupt a reader who thinks they know the categoryCorrection
Open cold, on social or in a first emailProvoke
Carry a positioning argument end to endTrap
Answer a reader who has already said no onceReframe
Give an unnamed problem a label that travelsNaming
Replace an adjective the reader cannot argue withSpecificity
Close a piece aimed at a buyer in an active evaluationSales Question

Everything the patterns operate on comes from the kit. Traps come from Named Traps in brand-kit/positioning.md, numbers from Headline Stats with their sources, and separating lines from Categories We Are Not. If you coin a new trap while rewriting, say so in the delivery note and tell the user to add it to Named Traps, so the name stays consistent across every asset. Do not leave a coined name in the copy without flagging it.

3.1 A note on external validation

Sourced third-party voices carry weight your own claims cannot. Use them verbatim, with attribution and a link, and only from External Validation Quotes in brand-kit/positioning.md. Never paraphrase a quote to make it fit, never quote someone who is not in the kit, and never invent an attribution. A quote you cannot link is a quote you cannot use.


4. Prompt Block

When another tool or a teammate needs the style compressed into a prompt, use this. Fill the bracketed parts from the kit.

Write in our house style.
Direct and action-led. Active voice. No appositive phrases. No em-dashes.
Short sentences. Put the action early.
Never use the contrast formula ("Most companies do X. Winners do Y." or
"It's not about X, it's about Y.").
No "it's not just X, it's Y." No "in today's..." openers. No "let's dive in."
No triple parallel lists in every paragraph. No concluding summary paragraph.
Banned vocabulary: [paste the Banned Words and Phrases list from voice.md]

Anchor the content to [canonical noun from The Frame in positioning.md] and use
its canonical definition verbatim on first mention.

Our qualities are [names, internal use only]. They are qualities, not section
headers. Never write their names or the word "pillars" or "framework" into
customer-facing copy.

Use these named traps where they fit: [from Named Traps]
Use these taglines verbatim or not at all: [from Naming and Usage Rules]
Only use numbers that appear in [Headline Stats], with their sources.
Close with the CTA for this context from [Calls to Action].

5. The Quality Panel

The panel is the final gate before anything goes out. Six experts score the draft 0-100. Target is 90 or above aggregate. Maximum three rounds.

5.1 When to run it

  • Always in quality panel mode.
  • Automatically at the end of rewrite mode. Produce the rewrite, then append the one-line tail. If the aggregate is below 90, offer the full panel.
  • On request in audit mode, if the user wants a number after the violation list.

5.2 The six experts

#ExpertWeightLens
1Voice Match1.0xDoes this read like the house wrote it? Direct, action-led, short sentences, no hedging, no em-dashes, no appositives. Checks against every section of brand-kit/voice.md.
2Positioning Enforcer1.0xChecks the draft against brand-kit/positioning.md. Does it sit inside The Frame? Does every claim trace to a section of that file? Are the qualities in The Pillars conflated with each other, or worse, exposed by name in the copy? Are taglines used verbatim? Is any stat used that is not in Headline Stats with a source?
3Target Buyer1.0xReads as the tier-one buyer defined in brand-kit/icp.md: the ICP column of The Fit Matrix, at the seniority and company size it names. Would that person find this credible, specific, and worth their time, or does it read like vendor marketing aimed at no one?
4AI Humanizer1.5xScores conformance to the house style, which is what most machine output fails. Mandatory. Uses the rubric in The Humanizer Rubric. Below 70 forces a revision regardless of aggregate.
5Pattern Checker1.0xIs at least one pattern from Messaging Patterns present, a named trap, a reframe, a correction, a naming? Or does the piece fall back on generic explanation?
6Specificity Auditor1.0xAre adjectives replaced with numbers, role titles, and named systems? Are the numbers sourced? Anything left as "fast," "many," "expensive," or "modern" costs points.

Experts 2 and 3 are parameterized by the kit. If positioning.md or icp.md still has template placeholders, those two experts cannot score. Say so and stop rather than inventing a standard.

5.3 The loop

Round 1: Score all 6 experts -> aggregate (Humanizer x 1.5)
  |- >= 90 -> output PASS, show summary, done
  |- <  90 -> identify top 3 weaknesses -> AUTO-REVISE -> Round 2

Round 2: Re-score the revised draft
  |- >= 90 -> output PASS with the iteration trail, done
  |- <  90 -> STOP and ASK: "Round 2 scored [X]/100. Top weaknesses: [list].
              Want me to run round 3, take the best version as-is,
              or change approach?"

Round 3 (only if the user approves): final attempt, then output the best
version with an honest score. Never run a fourth round.

Hard rule: if the Humanizer scores below 70, revise regardless of the aggregate. Copy that sounds generated is worse than no copy.

5.4 Aggregate math

weighted_total = (Voice + Positioning + Buyer + Pattern + Specificity)
                 + (Humanizer x 1.5)
divisor        = 5 + 1.5 = 6.5
aggregate      = round(weighted_total / 6.5)

If Humanizer < 70, force revision regardless of aggregate.

5.5 What to revise on

When the aggregate is below 90, take the three lowest raw scores and address each one:

  • Voice Match low? Cut em-dashes, kill hedging, shorten sentences, lead with the verb, split paragraphs over four sentences.
  • Positioning Enforcer low? Re-read brand-kit/positioning.md. Confirm the canonical noun and definition are present. Confirm each claim maps to the right quality and that no quality's name, "pillars," or "framework" appears in the copy. Strip any stat missing from Headline Stats.
  • Target Buyer low? Add role titles from Buying Committee, name real systems, add real metrics. Cut every sentence that could appear in any competitor's blog unchanged.
  • Humanizer low? Sweep the banned vocabulary first, then work the pattern table in order.
  • Pattern Checker low? Apply Trap, Reframe, Correction, or Naming. At least one must be present.
  • Specificity Auditor low? Replace every adjective of degree with a number, a role title, or a named system.

6. The Humanizer Rubric

Scott King wrote the 23 patterns below by hand while editing AI-drafted B2B marketing copy against his own published articles, naming each failure as it recurred. The weights measure damage to credibility with a technical reader rather than how often a pattern appears, which is why scaffolding exposure costs 10 and copula avoidance costs 5. docs/voice-calibration.md carries the reasoning and the method for recalibrating these weights against your own audience.

This is a house-style conformance score, not an AI detector. It is hand-built and deliberately opinionated, and it catches most machine output as a side effect, because machine output is unstyled by default. It does not identify machine authorship. A human draft written outside this house style scores low, and a machine draft edited into the house style scores high. Say "off style," never "written by AI."

Start at 100 and deduct per pattern detected. The full arithmetic, including how repeated patterns stack, is in 6.3 Scoring and reporting. Read it before you score anything.

6.1 Banned vocabulary, minus 5 per distinct term, capped at minus 25

Deduct 5 for every distinct offending term, drawn from the Banned Words and Phrases section of brand-kit/voice.md plus every term below. Count a term once no matter how many times it repeats: three uses of "leverage" cost 5, not 15. The two lists overlap on purpose, so a term sitting on both lists still counts once.

Cap the total vocabulary deduction at minus 25, however many distinct terms appear. The cap exists because the weighted patterns in Patterns to flag are tuned relative to each other, and an uncapped word count overrides them.

delve, tapestry, landscape (abstract use), leverage, multifaceted, nuanced, pivotal, realm, robust, seamless, testament, transformative, underscore (verb), utilize, whilst, keen, embark, comprehensive, intricate, commendable, meticulous, paramount, groundbreaking, innovative, cutting-edge, synergy, holistic, paradigm, ecosystem, Additionally (as a paragraph starter), align with, crucial, enduring, enhance, fostering, garner, highlight (verb), interplay, intricacies, showcase, vibrant, valuable, profound, renowned, breathtaking, nestled, stunning, dive into, game-changer, unlock

6.2 Patterns to flag

#PatternPenaltyTell
1Significance inflation-10"stands as," "is a testament," "pivotal moment," "underscores its importance," "marks a turning point"
2Vague attributions-8"Experts believe," "industry reports," "studies show," "many companies are saying", with no named source. Every claim of this shape needs an entry in Headline Stats or External Validation Quotes.
3Superficial -ing analyses-8Tacked-on "highlighting," "showcasing," "underscoring," "reflecting," "ensuring," "fostering" clauses that add nothing to the sentence
4Promotional language-8"boasts," "vibrant," "rich" (figurative), "in the heart of," "must-have," "world-class"
5Formulaic challenges/future-10"Despite these challenges, X continues to..." or "The future looks bright for..."
6Copula avoidance-5"serves as," "stands as," "represents," "marks," "boasts," "features", when "is" or "has" would do
7Negative parallelism-5"It's not just X, it's Y." Banned in brand-kit/voice.md; double-flag it here
8Rule-of-three overuse-8Every paragraph forced into triple adjectives, triple nouns, or triple parallel clauses
9Elegant variation-5Synonym cycling for the same noun: "the CEO... the executive... the business leader..."
10False ranges-5"From X to Y" where X and Y are not on the same scale: "from content to compliance, from ROI to ethics"
11Em-dash overuse-5More than one em-dash per 200 words. Em-dashes are banned outright in brand-kit/voice.md; penalize on sight here
12Excessive hedging-8"could potentially," "might possibly," "may perhaps," "it could be argued," "can help"
13Filler phrases-5 each"in order to" (use "to"), "due to the fact that" (use "because"), "at this point in time" (use "now"), "it is important to note that" (just state it)
14Generic positive conclusions-10"The future looks bright," "exciting times ahead," "stay tuned," "the possibilities are endless"
15Sycophantic openers-8"Great question!" "You're absolutely right!" "That's an excellent point!"
16Knowledge-cutoff disclaimers-10"As of [date]," "while specific details are limited," "based on available information"
17Collaborative artifacts-10"I hope this helps," "Of course!" "Certainly!" "Let me know if you'd like...", assistant chatter that survived into the asset
18"In today's..." openers-10"In today's fast-paced world," "in the current landscape," "in an era of..."
19The contrast formula-10"Most companies do X. Winners do Y." and "It's not about X, it's about Y." Structurally hollow. Never write one, never prescribe one
20Dive/unpack/imagine openers-8"Let's dive in," "let's unpack," "picture this," "imagine if"
21Concluding summaries-8A final paragraph that restates what was just said. End on the CTA, a question, or the sharpest line, never on a recap
22Question-as-opening-5Repeating the user's question back as the first line, or a rhetorical "Ever wonder why...?" hook
23Scaffolding exposure-10Your own internal structure surfacing in customer-facing copy: your qualities named as a set, "our pillars," "our framework," or your framework used as section headers. See the scope note below before firing this. Automatic revision trigger

Scope note on pattern 23. This pattern fires only on the writer's own scaffolding: the qualities in The Pillars, the frame in The Frame, or any internal messaging structure from brand-kit/positioning.md appearing by name in copy meant for a reader. It does not fire on a framework the piece is about.

  • Fires: "Our three pillars are speed, trust, and scale." The reader is being shown the scaffolding behind the writing.
  • Does not fire: "The four pillars of agentic governance are identity, audit, approval, and spend control." That is the subject matter. A piece is allowed to organize an argument.

The test is ownership, not the word "pillar." If the structure belongs to the argument, it is content. If it belongs to how you decided what to say, it is scaffolding. At minus 10 with an automatic revision trigger attached, firing this one wrongly is expensive: in the calibration study a scorer who applied it to subject-matter frameworks scored one piece 74 points below a scorer who did not.

6.3 Scoring and reporting

  • 90-100: Reads like the house wrote it. Ship it.
  • 70-89: Minor tells. Fixable in one revision pass.
  • 50-69: Obvious pattern damage. Significant rewrite needed.
  • 0-49: Off style entirely. Rewrite from the source idea, not from the draft.

Stacking. A pattern that fires more than once deducts up to twice its base penalty, and no further. A pattern firing eight times costs the same as one firing twice. Apply stacking first, then the vocabulary cap, then clamp. This rule is easy to skip and it materially changes a score: in the calibration study, a scorer who missed it ranked the same nine pieces in a different order.

Compute the arithmetic in full, then clamp. Total every deduction first, the capped vocabulary penalty plus every weighted pattern, subtract it from 100, and only then clamp the result into the 0 to 100 range. Never clamp an intermediate step. A draft carrying 92 points of deductions scores 8, not 0.

Report deduction density. Alongside the score, give the total deduction per thousand words. Distinct patterns accumulate with length even though each one stacks no higher than twice its base, so a long piece trips more of them than a short one on the same prose habits. The density figure is diagnostic only and does not change the score. Never compare the raw scores of two pieces of very different length without it.

Report both numbers. Give the clamped score and the raw total beside it: "Humanizer: 0/100, raw minus 31, total deduction minus 131." A draft at a raw minus 31 and a draft at a raw minus 5 both clamp to 0, and the clamped number on its own hides the 26 points between them.

Use the band, not the number. The band and the ranking are what this instrument produces reliably. The raw score is a diagnostic that travels with them, and it moves from reader to reader more than the band does.

  • The band and the relative ranking are the reliable outputs. The raw score is diagnostic: it says how much damage the patterns did, not what the piece is worth to the point.
  • Never report a score to a third party as a precise measurement. Report the band, and report the specific patterns that fired, which is the actionable part anyway. "Scored 72" claims a precision this rubric does not have.
  • Two scorers reading this rubric on the same draft can differ by twenty points or more. That is a property of the instrument, not a failure of either reader.
  • A score within ten points of a band boundary decides nothing until a second independent pass agrees with it. Band edges are exactly where a difference in scorer severity flips the verdict.
  • Patterns 1 and 5 carry the loosest tells on the table and are where severity diverges most. Fire them only when the tell is unmistakable, never when it is merely arguable.

7. Output Formats

7.1 Quality panel mode

## Result: [SCORE]/100, [PASS | NEEDS WORK]

[Final draft here]

**Iterations:** [N] round(s)
**Panel:** Voice Match, Positioning Enforcer, Target Buyer,
AI Humanizer (1.5x), Pattern Checker, Specificity Auditor

---

<details>
<summary>Scoring rounds</summary>

### Round 1, Aggregate: [X]/100

| Expert | Score | Key feedback |
|---|---|---|
| Voice Match | 87 | Two em-dashes in paragraph 3, hedging in the opener |
| Positioning Enforcer | 92 | Canonical definition present and verbatim |
| Target Buyer | 78 | No role named. Reads as marketing, not analysis |
| AI Humanizer (1.5x) | 81, raw minus 19 | "leverage" once, "robust" once, one significance-inflation tell |
| Pattern Checker | 65 | No named trap. Generic explanation throughout |
| Specificity Auditor | 88 | Good numbers mid-piece, vague closer |

**Top 3 weaknesses:** [ranked]
**Changes made:** [specific edits]

### Round 2, Aggregate: [X]/100
[same format]

</details>

7.2 Rewrite mode tail

After producing the rewrite, append one line:

---
**Quality check:** [SCORE]/100, Voice [X], Positioning [X], Buyer [X],
Humanizer [X] (raw [Y]), Pattern [X], Specificity [X]. [Pass | Want the full
panel?]

Below 90, offer the panel. At 90 or above, ship the rewrite as-is.

7.3 Audit mode

A table of violations, most severe first: the quoted text, which rule it breaks, which brand-kit heading covers it, and the specific replacement. No score unless the user asks.


8. Cross-References

All optional. This skill works alone.

  • brand-kit-setup: run it if the kit is missing or still has placeholders.
  • content-writer: if it is installed, use it to generate a long-form asset, then run this skill as the final pass.
  • social-generator: if it is installed, prefer it for short social posts; this skill still scores the output.
  • transcript-to-article: if it is installed, it produces the draft and this skill grades it.

Anything not in skills/ does not exist. Do not reference it.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.