writing-clone-starter

Writing Clone Starter

What this is

This is not an override modification of writing-clone-profile or writing-clone-lite .

This is a new user entry point, specifically handling two types of scenarios:

1. The user does not have their own representative works, memory assets, or long-term accumulation, but wants to first write a strong article
2. The user wants to explicitly switch to a certain built-in author direction and first get a high-mimicry experience

In one sentence:

What this skill solves is "write it first," not "first prove that you already have complete personal style assets."

Current structure

starter currently defaults to only two parts:

1. the skill itself：.claude/skills/writing-clone-starter/
2. unified material library：02_素材库/writing-clone-starter-material-library/

Boundary requirements:

• starter internal rules, modes, and author assets continue to stay within the skill
• All external materials, evidence, clippings, and verification materials outside the skill are uniformly assigned to the material library
• starter no longer by default reaches across to inbox, project research areas, or other scattered system paths to fetch material

When to use

When the following signals appear, prioritize this skill:

• The user has only one topic, one direction, or one rough idea, but wants to first write it into a strong draft
• The user explicitly says they do not have much personal experience, material, or memory accumulation to draw on
• The user wants to first experience the system's capabilities with a low barrier to entry
• The user explicitly says they want to "write in the style of Dan Koe"
• The user explicitly says they want to "write in the style of Zhou Zuoluo"
• The user explicitly says they want to "write in the style of Justin Welsh / Liu Run / Liang Ning / Xue Hui"
• What the user wants is "first write a strong draft," not "first write like myself"

When not to use

When the following signals appear, do not use this skill:

• The user already has mature personal style assets and wants to continue writing like themselves
• The user is already in the existing process of writing-clone-profile / writing-clone-lite in the existing workflow of
• The current task is not writing an article, but continuing to research the system or expand materials

In this case, prioritize returning to:

• writing-clone-profile
• writing-clone-lite
• or the corresponding research / material skill

Currently implemented scope

Currently three layers of assets have been implemented:

• references/evaluation/: common standards
• references/content-archetypes/: first batch of 4 prototypes for strong article mode
• references/built-in-profiles/dankoe/: prototype of Dan Koe's formal author profile
• references/built-in-profiles/zhouzuoluo/: prototype of Zhou Zuoluo's formal author profile
• references/built-in-profiles/justinwelsh/: prototype of Justin Welsh's formal author profile
• references/built-in-profiles/liurun/: prototype of Liu Run's formal author profile
• references/built-in-profiles/liangning/: prototype of Liang Ning's formal author profile
• references/built-in-profiles/xuehui/: prototype of Xue Hui's formal author profile

Maturity notes:

• dankoe/、zhouzuoluo/: the evidence layer is relatively thicker, and they are the more mature pilot authors in the current starter
• justinwelsh/: has reached the trial-ready draft level within the current starter architecture
• liurun/、liangning/、xuehui/: currently still belongs to narrow bandwidth formal-profile-draft, more suitable for trial within their respective core bandwidths, and should not by default be understood as having the same maturity as the first two

and one more layer has been added:

• references/runtime-closure.md: unified run order and output contract
• references/profile-distillation/: built-in persona profile distillation module (maintainer path)

The complete action library and full runtime automation are still out of scope for this round.

Currently not implemented scope

This version explicitlydoes not yet have：

• Complete action library compendium
• A larger-scale multi-author profile pool
• Full runtime automatic routing implementation
• Benchmark system upgrade

In other words, this version has already put the boundaries, standards, the first version of the strong-article prototype, and the first batch of built-in author profiles locally, but it is not yet a full-featured version.

Mode definitions

Mode A: Strong Article Mode

• Default mode
• Not bound to an author
• Prioritizes answering: how can this article most easily be written strong, complete, and publishable
• It is the default main working path of the starter
• At the current stage, the shared standard is used as the acceptance baseline

Mode B: High-Mimicry Mode

• Entered only when the user explicitly specifies an author
• Prioritizes answering: how can this article be as much like the specified author themselves as possible
• Currently built-in authors: Dan Koe, Zhou Zuoluo, Justin Welsh, Liu Run, Liang Ning, Xue Hui
• If the topic exceeds the author's native topic bandwidth, you must warn and automatically downgrade to Strong Article Mode

Unified runtime decision order

By default, the runtime processes in the following order; do not skip steps:

1. First check whether the user explicitly specifies an author
2. If an author is specified, first try to enter High-Mimicry Mode
3. Before entering high mimicry, a bandwidth judgment must first be made
4. If weak bandwidth is hit or a downgrade condition is triggered, switch directly back to Strong Article Mode
5. If no author is specified, go directly to Strong Article Mode
6. After entering Strong Article Mode, then determine the prototype
7. After hitting the prototype, organize the body according to the material selection rules and anti-bleed constraints

For a more detailed operating protocol, see:references/runtime-closure.md

Minimal routing principle

Default routing

• The user has not specified an author: by default, according toStrong Article Modeunderstand
• The user explicitly specifies Dan Koe: enterHigh-Mimicry Mode
• The user explicitly specifies Zhou Zuoluo: enterHigh-Mimicry Mode
• The user explicitly specifies Justin Welsh: enterHigh-Mimicry Mode
• The user explicitly specifies Liu Run: enterHigh-Mimicry Mode
• The user explicitly specifies Liang Ning: enterHigh-Mimicry Mode
• The user explicitly specifies Xue Hui: enterHigh-Mimicry Mode

Among these, keep in mind:

• Strong Article Mode is not a downgrade fallback, but the default main path
• High mimicry is an explicit branch, not a default parallel main entry point

Downgrade principle

If the topic clearly exceeds the specified author's native topic bandwidth:

1. Clearly warn
2. Automatically switch back to Strong Article Mode
3. Do not force likeness, and do not output a fake mimicry draft

Article-level output contract

Delivery priority

Deliver first by default:

• response.md: a finished body text that can be directly published or directly evaluated

Explanatory material is placed only in:

• run_summary.md: record the mode selection, bandwidth judgment, reading basis, and degradation reason

Do not, in turn, let run_summary.md reflect mode differences more than response.md .

The boundary between the body text and the explanation

By default, do not write the following into the body text:

• "I am now switching to a certain mode according to the skill rules"
• "Because topic-bandwidth stipulates this, therefore..."
• "Which profile assets I read"
• any system note-style explanation

By default, degradation reminders should not enter response.mdeither. Prefer to leave the degradation reason in run_summary.md。

Only when the task itself explicitly requires "explain to the user why it did not continue writing in that author's style" is a separate explanation outside the body text allowed; the default file deliverable should still be a clean final draft. The body text should start directly from the title or the first paragraph of the body, not from a system explanation.

Mode differences must land at the body text level

Final acceptance should not only look at run_summary.md。

The more important question is:

Looking only at response.md, can one feel which type of mode this draft follows.

If the body text barely shows any mode differences and one can only rely on the explanation file to explain why they differ, then the output contract has not truly been established.

References reading order

Strong article mode

1. references/runtime-closure.md
2. references/evaluation/strong-article-rubric.md
3. references/archetype-router.md
4. references/content-archetypes/<命中的原型>.md
5. references/library-first-retrieval.md
6. references/archetype-anti-bleed.md
7. The user's current topic and constraints

High mimicry mode (specified author)

1. references/runtime-closure.md
2. references/evaluation/high-likeness-rubric.md
3. references/built-in-profiles/<指定作者>/profile-card.md
4. references/built-in-profiles/<指定作者>/worldview.md
5. references/built-in-profiles/<指定作者>/recurring-moves.md
6. references/built-in-profiles/<指定作者>/taboo-moves.md
7. references/built-in-profiles/<指定作者>/snippet-bank.md
8. references/built-in-profiles/<指定作者>/topic-bandwidth.md
9. references/built-in-profiles/<指定作者>/hard-signals.md
10. references/built-in-profiles/<指定作者>/contrastive-pairs.md
11. references/built-in-profiles/<指定作者>/false-positive-signals.md
12. references/built-in-profiles/<指定作者>/acceptance-checklist.md
13. references/built-in-profiles/<指定作者>/raw-corpus/README.md

Principles:

• Strong article mode reads the standards first, then determines the archetype, then gathers material
• High mimicry mode reads the standards first, then reads the author assets
• All modes share the same operating order; do not improvise separately
• Do not read form-normalized-anchors/
• Do not mix in old personal profile assets

Currently available assets

• Runtime protocol:references/runtime-closure.md
• Strong article standards:references/evaluation/strong-article-rubric.md
• Strong article archetypes:references/content-archetypes/
• Strong article routing:references/archetype-router.md
• Strong article material gathering:references/library-first-retrieval.md
• Strong article flavor-mixing prevention:references/archetype-anti-bleed.md
• Strong article minimum test set:references/archetype-evals.md
• High mimicry standards:references/evaluation/high-likeness-rubric.md
• Dan Koe profile：references/built-in-profiles/dankoe/
• Zhou Zuoluo profile:references/built-in-profiles/zhouzuoluo/
• Justin Welsh profile：references/built-in-profiles/justinwelsh/
• Liu Run profile:references/built-in-profiles/liurun/
• Liang Ning profile:references/built-in-profiles/liangning/
• Xue Hui profile:references/built-in-profiles/xuehui/
• Unified material library:02_素材库/writing-clone-starter-material-library/

Progressive Disclosure

Do not move all the ideas of the entire skill into context right away.

The reading principle of the current version is very simple:

• Read only the few documents that the current mode truly needs
• Read runtime-closure first, then read the assets corresponding to the current mode
• Strong article mode first grasps the common standards, then grasps the archetype-related files
• Only high mimicry mode goes to read the corresponding author's profile assets
• For modules that are not currently implemented, do not pretend they already exist

Built-in maintenance modules (not a regular user entry point)

The starter currently also has a set of internal reference files built in that serve only maintainers:

• references/profile-distillation/
• references/profile-probe/

What it solves is:

• Adding a new built-in author profile
• Updating an existing author profile
• Auditing the evidence quality of a certain author profile
• Verifying whether a certain built-in profile can truly produce drafts stably
• Use blind probe / cold review to judge whether a profile reaches stable 90 readiness

Itis notthe default trigger path for ordinary learners.

Do not switch into this module when the user just wants to "write an article."

Among these, distinguish:

• profile-distillation/: generate profile
• profile-probe/: validate profile

Mode layering requirements (surface plane)

Strong article mode

The body text goals are:

• More like a high-quality strong Chinese draft
• Explain the problem clearly and push it toward a judgment or action
• For the same topic with different archetypes, the body skeleton must be clearly distinguishable
• Do not by default borrow Dan Koe's author signal package to manufacture a sense of sophistication

High mimicry mode

The body text goals are:

• Make the author's action combinations appear in the body text
• Not just the appearance of public words like "system / map / sovereignty / compounding"
• Truly embody the opening move, the redefinition path, the abstraction-lifting method, and the closing convergence method
• Let at least one complete author action chain run through the body text, rather than scattering author signals in a fragmented way

Downgraded strong article

If downgrade is triggered:

• The body text must still read like a complete finished draft
• By default, do not write the internal decision process into response.md
• By default, leave the downgrade reason in run_summary.md
• The win point should be "more stable, more publishable," not "better at explaining why it downgraded itself"

Current hard constraints

1. Do not read form-normalized-anchors/
2. Do not require the user to initialize their own profile first
3. Do not treat user memory as a default premise
4. In high mimicry mode, author assets take priority over the general material library
5. Once it exceeds the author's topic bandwidth, downgrade directly; do not force a fake resemblance

Responsibility boundaries at the current stage

The responsibility of this version of the skill is to take:

• runtime closure protocol
• common standards
• the first batch of strong article archetype libraries
• dual-mode boundaries
• the prototypes of formal author profiles for Dan Koe, Zhou Zuoluo, Justin Welsh, Liu Run, Liang Ning, and Xue Hui

and organize them into a more formal, more triggerable, more evaluable entry point.

It currentlyis not responsible for：

• complete feature implementation
• parallel invocation of multiple authors
• deep runtime integration with the old skill