SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsenterprise-clone-builder

enterprise-clone-builder

General & other

View repository

Updated 10-16 04:57

About this skill

Enterprise Clone Builder

What is this

The repository building tool for enterprise clones. Input the enterprise name and materials, output a complete, standardized clone repository.

In one sentence:builder builds the repository, writer writes. The repository produced by builder is directly consumed by enterprise-clone-writer.

Users

Zhijian AI delivery team (Dapeng or team members). The users are professionals, not enterprise clients.

What clients buy is the finished clone product. builder is a production tool, not a self-service tool delivered to clients.

Input parameters

Parameter Required Description
Enterprise name Yes Full name, used to create the repository directory and for online search
Official website URL No If available, scrape the official website; if not, skip
Local materials path No Path to the materials folder provided by the client
Output path No By default, create in the current directory {企业名}-企业分身/

If there is a local materials path,local materials take priority over online research—the best materials are always local to the enterprise.

Main process

Step 0: Initialization + local materials inventory

0.1 Create the standard directory structure

Create the repository root directory and all subdirectories based on the enterprise name. For directory specifications, see references/directory-spec.md。

0.2 Scan local materials

If a local materials path is provided:

  1. Recursively scan all files under that directory
  2. Identify file formats (PDF / Word / Excel / PPT / images / Markdown / plain text)
  3. Read the first 500 characters of readable files to determine the content type
  4. Archive by content type into 02-原始素材/本地资料/ under the corresponding subdirectory

For intelligent classification rules, see references/local-intake-guide.md。

Files whose type cannot be determined go into 本地资料/未分类/, and are marked in the inventory report.

0.3 Diagnose coverage

After scanning all files, output a coverage report for 7 dimensions:

Dimension Judgment criteria
Sufficient More than 3 files
Insufficient 1-2 files
Missing 0 files

0.4 Output the materials inventory report

Under 05-调研记录/ , generate 资料盘点报告.md, recording the scan results, archiving status, and gap diagnosis.

If --interactive mode, stop here and let the user confirm or supplement before continuing.

Step 1: Online supplementary research

Only fill the gaps diagnosed in Step 0, do not scrape everything.

  1. Use web-clipper to scrape the official website (if there is a URL)
    • Official website homepage
    • Product list page → extract sublinks from it → scrape representative product details (10-20)
    • News list page → extract sub-links from it → scrape representative news details (10-20 items)
  2. Search for the company name and scrape third-party platform information
  3. Fill gaps in a targeted way: insufficient external communication → focus on scraping news; insufficient products → focus on scraping product pages
  4. Register all scraped files to 02-原始素材/source-map.md

If the company is too small and there is almost no information on the internet: mark "insufficient public information, avatar quality depends on local materials" and continue.

For how to call web-clipper, see references/web-clipper-usage.md。

Step 2: Structured extraction

Extract from all materials (local + online) and write into the company profile and content assets.

2.1 Company profile: 5 dimensions → 01-企业画像/

File Extracted content
01-Basic Information.md Business registration, scale, address, contact information
02-Products and Services.md Product line classification, core technologies, application scenarios
03-Customers and Market.md Target customers, industry distribution, competitive landscape
04-Brand Tone.md Brand voice, expression style, prohibited expressions
05-Development History.md Founding time, key milestones

2.2 Content assets: 4 libraries → 03-内容资产/

File Extracted content
Product Description Library.md Standardized description for each product (model/parameters/use cases/selling points/source)
Customer Case Library.md Customer cases and application scenarios
Technical Knowledge Library.md Industry background, technical principles, standards
Brand Materials.md slogan, introduction, standard copy, marketing talking points

2.3 Source annotation

Annotate the source file for all extracted information. Distinguish between "evidenced" and "inferred". Write to 05-调研记录/claims-ledger.md。

For detailed extraction specifications, see references/extraction-guide.md。

Step 3: Writing style analysis

3.1 Select representative articles

Select 3-5 representative articles from external communication materials. Priority: local WeChat official account articles > official website news > third-party platforms.

For selection criteria, see references/voice-analysis-guide.md。

3.2 Distill writing style features → 04-文风样本/voice-cues.md

Extraction dimensions: tone and persona, self-reference style, sentence patterns and rhythm, word choice features, structural preferences, product description style, prohibited expressions. Attach original text evidence to each item.

3.3 Normalized sample articles → 04-文风样本/form-normalized-anchors/

Select 1-2 articles for normalization (clean up formatting, unify heading levels).

3.4 Original sample article anchors → 04-文风样本/voice-anchors.md

Include the full text or key paragraphs of 3-5 representative articles.

3.5 Writing style confidence

When there are fewer than 3 external communication materials, mark in voice-cues.md: "low writing style confidence, inferred based on limited samples".

Step 4: Build the library

4.1 AGENTS.md

Generate the avatar's system prompt based on the company profile and brand tone. Include: identity, minimal company profile, core product lines, expression principles, work rules (read paths, quality bottom line, output archiving specifications), quick routing.

4.2 00-Company Overview.md

One-page quick reference: basic information table, product line overview, customers and market, multi-platform presence, key development milestones, writing style quick reference, factual red lines.

4.3 README.md

Repository description + directory structure + usage guide + template reuse instructions.

4.4 Three-Piece Research Record Set → 05-调研记录/

  • claims-ledger.md: Sources and confidence levels for all factual assertions
  • gap-report.md: Resolved gaps + remaining gaps
  • 调研日志.md: Knowledge base creation process record

4.5 source-map.md → 02-原始素材/source-map.md

Source map of all materials (local + online).

Completion Criteria

The knowledge base is considered complete when all of the following files exist and are non-empty:

AGENTS.md                          ✓
00-企业概览.md                      ✓
01-企业画像/01-基本信息.md          ✓
01-企业画像/02-产品与服务.md        ✓
01-企业画像/03-客户与市场.md        ✓
01-企业画像/04-品牌调性.md          ✓
01-企业画像/05-发展历程.md          ✓
03-内容资产/产品描述库.md           ✓
03-内容资产/客户案例库.md           ✓
03-内容资产/技术知识库.md           ✓
03-内容资产/品牌素材.md             ✓
04-文风样本/voice-anchors.md        ✓
04-文风样本/voice-cues.md           ✓
05-调研记录/claims-ledger.md        ✓
05-调研记录/gap-report.md           ✓
05-调研记录/调研日志.md             ✓
02-原始素材/source-map.md           ✓
06-产出/                            ✓（空目录，留给writer）

README.md is optional (can be added later if time is tight).

Run Modes

Default: Fully automatic

Run through Step 0 → 4 in one go, and finally output the complete repository + material inventory report.

Segmented mode:--interactive

Stop and report at the end of each step:

  • After Step 0: Show the material coverage report and confirm gaps
  • After Step 1: Show the scraping results and material volume
  • After Step 2: Show the profile key points and confirm accuracy
  • After Step 3: Show the writing style feature summary

Continue to the next step after user confirmation.

Use subagents to speed up

Step 1 (scraping) and Step 2 (extraction) of this skill are suitable for parallel execution with subagents:

  • After Step 0 is complete, Step 1 can be split into multiple subagents to scrape different sources simultaneously
  • After Step 1 is complete, Step 2's enterprise profile extraction, content asset filling, and writing style analysis can run in parallel
  • Step 4's AGENTS.md, overview, and quality files can also be written in parallel

Practical advice: 3-4 subagents running in parallel can compress knowledge base creation time from most of a day to 1-2 hours.

Integration with enterprise-clone-writer

The repository produced by builder is the reading convention for writer:

writer reads builder produces
AGENTS.md Step 4.1
01-Enterprise Profile/04-Brand Tone.md Step 2.1
04-Writing Style Samples/voice-cues.md Step 3.2
03-Content Assets/*.md Step 2.2

The two are loosely coupled through the directory structure convention. writer does not directly call builder.

Applicable Boundaries

Suitable for：

  • Enterprises with a certain amount of historical accumulation (3+ years)
  • Enterprises with local material accumulation (product docs, historical content, customer cases, etc.)
  • Enterprises that have not yet systematized digitalization

Not suitable for：

  • Small enterprises with no local materials and almost no information on the internet
  • Enterprises that have just been founded and have no material accumulation

Not doing

  • Not doing writing (leave it to enterprise-clone-writer)
  • Not doing PPT generation (leave it to zhijian-ai-visual-deck)
  • Not doing customer delivery packaging
  • Not doing digital human videos
  • Not guiding AI beginner users

File Responsibilities

  • SKILL.md(this file): Main entry point, process, inputs, outputs, completion criteria
  • references/directory-spec.md: Standard directory structure specification
  • references/local-intake-guide.md: Intelligent classification rules for local materials + customer collection checklist
  • references/extraction-guide.md: Structured extraction specification
  • references/voice-analysis-guide.md: Writing style analysis operation guide
  • references/web-clipper-usage.md: web-clipper invocation method

Principles:SKILL.md is authoritative. references are expanded explanations.

Customer Material Collection Checklist

Before building the knowledge base, you can export the collection checklist in references/local-intake-guide.md and send it to the customer, so the customer can organize the material folder according to the checklist in advance.

The checklist includes 7 categories of materials (company introduction/product materials/customer cases/external communications/sales services/industry knowledge/brand visuals), and each category notes what documents are needed, examples, and priority.

After the customer finishes organizing the materials, they just need to point to that folder to launch the skill.

Last updated: 2026-07-08