The long-awaited Windows version of the WeChat Skill is finally done. Everyone, please keep using it quietly — don’t spread the word, I really don’t want to receive another lawyer’s letter. Thanks, everyone! Full analysis: chats, Moments, Favorites, contacts. Skill: https://github.com/mcncarl/yichen-skills/blob/main/yichen-wechat-local-vault/README.md

mcncarl/yichen-skills

Source: https://github.com/mcncarl/yichen-skills

yichen-skills

English | 中文

A skill collection for creators who want to streamline writing, X content slicing, X Articles draft publishing, WeChat digital-asset capture, and local workflows with Claude Code and Codex.

Maintainer

Created and maintained by 逸尘 (Yichen), a serial entrepreneur born in 2006, law student, and founder of Yu Hackathon (渝客松), working on AI products, content, and services for businesses.

• Personal website: yichen.ai (https://yichen.ai/) · About me (https://yichen.ai/about/)
• Social media: X (https://x.com/gengdaJ) · YouTube (https://www.youtube.com/@yichenai) · Douyin (https://v.douyin.com/diF63UgPZtI) · Xiaohongshu (https://xhslink.com/m/3eiF47bau17)
• WeChat: yichen10801; backup: yichen365ai
• When adding me, please include GitHub in the verification message

See yichen.ai (https://yichen.ai/) for current personal information, products, services, and contact details.

Personal Use and Commercial Authorization

This repository is available only for personal learning and non-commercial personal use. Commercial use—including client delivery, paid products or services, internal company deployment, marketplace packaging, and course bundling—requires the author’s prior explicit written authorization. To request commercial authorization, contact me on WeChat at yichen10801 and include Commercial Authorization in the verification message. Sending a contact request does not itself grant permission; commercial use is allowed only after you receive explicit written authorization.

Join TradeWinds, Yichen’s paid community, to unlock commercial authorization for all Skills in this repository and connect with more peers born in 2005 or later and experienced practitioners with proven results. Contact Yichen on WeChat (yichen10801) for membership pricing. The specific scope is subject to written authorization; third-party components remain governed by their original licenses.

What This Repo Does

1. Upload Obsidian/Markdown articles to X Articles drafts (yichen-x-article-draft-uploader)
2. Run two WeChat accounts on one Mac with a distinct blue icon (yichen-mac-wechat-dual-open)
3. Turn local WeChat chats, Moments, and Favorites into digital assets on macOS and Windows (yichen-wechat-local-vault)
4. Fetch benchmark videos from known Douyin links (yichen-content-archive)
5. Fetch benchmark posts from known Xiaohongshu links (yichen-content-archive)
6. Transcribe, caption, and rough-cut talking-head videos with Volcengine ASR (yichen-volc-asr)
7. Run verified research through the official ChatGPT web page (yichen-chatgpt-web-research)
8. Install and maintain a Markdown/Obsidian-first Agent Memory Vault system (yichen-agent-memory)
9. Batch-export WeChat Official Account article history, original-article lists, bodies, and optional read/comment metrics (yichen-wechat-mp-batch-exporter)
10. Read and export local WeCom/企业微信 5.x database snapshots without controlling the app (yichen-wecom-local-vault)
11. Let GPT call Grok for native X search or an independent second opinion without switching the main model (yichen-grok-consult)
12. Route multi-stage internet research through one safety-first entry point (yichen-web-research)
13. Normalize public web and platform search into reviewable candidates (yichen-unified-search)
14. Read, download, and archive only known or explicitly confirmed links (yichen-content-archive)
15. Export private bookmarks directly behind current-task authorization (yichen-bookmarks-export)
16. Choose between Step and Doubao/Volcengine ASR without duplicate submissions (yichen-asr)
17. Create authorized WeCom documents and manage todos, meetings, and schedules through the official CLI without controlling the desktop app (yichen-wecom-operations)
18. Turn one public X Post or Thread URL into verified 3:4 image slices and a finished video that embeds complete native video visuals and preserves their original audio when present (yichen-x-slicer)
19. Orchestrate ChatGPT Pro research, architecture, and read-only project review while Codex alone edits and tests (codex-chatgpt)

Included Skills

yichen-x-article-draft-uploader

Upload Obsidian/Markdown long-form articles to X Articles drafts:

• Uses a leading image as the optional 5:2 cover; otherwise keeps the draft cover empty
• Converts Markdown into rich text for the X editor
• Converts supported pipe tables into native X table blocks
• Inserts up to 25 body media items at their original Markdown positions
• Runs in an independent Playwright browser so it does not take over the user’s current Chrome window
• Imports Chrome login state into a private cookie file without committing it to the repository
• Reloads the same draft and verifies text, tables, media identity, count, order, and position
• Saves drafts only and never clicks the final 发布 button

See yichen-x-article-draft-uploader/README.md for installation, privacy notes, and troubleshooting.

yichen-mac-wechat-dual-open

Run two WeChat accounts simultaneously on macOS — no third-party tools:

• Copies WeChat, changes the bundle identifier, and re-signs locally
• Recolors the second app’s icon from green to blue for visual distinction
• Handles both outer and embedded icon files, Finder custom icon, and cache refresh
• One-command workflow: create → recolor-icon → launch
• Typical triggers: “微信双开”, “WeChat dual open”, “double WeChat”
• Requirements: macOS 12+, WeChat at /Applications/WeChat.app, Python 3.10+, Pillow
• Limitations: breaks after WeChat updates (re-run repair), push notifications may be unreliable
• Based on the well-known copy + bundle-id + ad-hoc signing method documented by @koffuxu (https://x.com/koffuxu/status/2043110831584690427)

yichen-wechat-local-vault

Unified full local WeChat data parsing on macOS and Windows. Windows full-database capture, authenticated decryption, and row-by-row chat parsing are implemented; Favorites and Moments body parsing has also passed real-data acceptance through the bundled shared-function workflow.

• macOS: full/incremental decryption, chats, Moments, Favorites, digests, and missed/reply-pending checks
• Windows full local capture: windows capture --scope all --source-id SOURCE_ID covers all databases in the selected account’s local data directory, including Favorites, Moments, media, and search indexes; the default capture scope remains chat databases
• Windows chat queries: snapshot --snapshot <path> validates, queries, searches, and exports the captured plaintext snapshot. Complete parsing is verified by comparing source rows, parsed rows, and output readback; individual query/export limits are separate
• Windows Favorites/Moments bodies: use the existing pure functions and read-only snapshot workflow in the Windows guide; dedicated query subcommands are not yet packaged
• Real acceptance covers full local capture, chat parsing, Favorites/Moments bodies, newly arriving messages, and export readback on Windows 11 build 26300 ARM64 with AMD64 WeChat 4.1.15.56 and CPython 3.12.10 AMD64
• Scope: data already synced to the selected computer. Three known search-index integrity gaps are reported explicitly; media-file restoration, ordinary non-administrator execution, other client versions/architectures, and complete Mac workflow parity still require further work or verification
• Does not operate the WeChat UI; login and test messages are performed manually by the user. Windows keys remain in memory; plaintext snapshots stay in private local storage
• Mac dependencies: Python 3.9+, pycryptodome, zstandard; Windows uses CPython 3.12 AMD64, native Windows CNG, and the pinned zstandard dependency
• See the Skill README, Windows full local parsing guide, and offline snapshot guide

Social fetchers integrated into yichen-content-archive

The former standalone Douyin and Xiaohongshu fetchers now have one source of truth:

• douyin_download.py reads metadata or downloads a known Douyin video through Playwright interception
• xiaohongshu_fetch.py anonymously reads known posts first, then downloads requested video, subtitles, or images
• Existing outputs are never overwritten; a new -run-N path is selected instead
• Xiaohongshu cookies require explicit current-task authorization, and optional Feishu deposition remains opt-in

yichen-volc-asr

Transcribe local audio/video files and generate rough cuts:

• Uses environment variables for Volcengine ASR and TOS configuration
• Produces transcript text, SRT subtitles, ASR cache, and optional rough-cut MP4
• Requires explicit user approval before cleaning temporary files

yichen-chatgpt-web-research

Legacy research-only entry for the user’s already signed-in official ChatGPT website account:

• Uses the real ChatGPT web page, not the OpenAI API or a separate account
• Prefers Chrome extension control and falls back to visible Computer Use only when necessary
• Waits for a full answer with a unique marker before extracting
• Saves raw and readable Markdown reports under the current workspace’s reports/ directory
• Keeps profile names, local paths, cookies, tokens, and browser storage out of the public skill

See yichen-chatgpt-web-research/README.md for privacy notes and workflow details.

Unified entry: codex-chatgpt

Run a bounded Codex × ChatGPT Review Loop:

• ChatGPT Pro performs public-web research, architecture, PLAN, and final REVIEW
• Codex is the only local file writer and command/test executor
• Code, hybrid, and review modes require a separately configured Secure Tunnel and seven-tool read-only MCP
• Pure research never starts or attaches the code Tunnel
• The public package contains no Runtime Key, Tunnel/App ID, private runtime, browser session, screenshot, or personal absolute path

This is the unified research + architecture + review entry. The older yichen-chatgpt-web-research directory remains available as a legacy research-only workflow. See codex-chatgpt/README.md for installation and the external runtime contract.

yichen-agent-memory

Install and maintain the public Agent Memory Vault system:

• Creates a local Markdown/Obsidian-first memory vault from the public template
• Uses Markdown as the source of truth and SQLite/FTS as the fast index
• Supports optional Zvec semantic retrieval for fuzzy “meaning-based” recall
• Guides prewrite reconcile, closeout, audit, and privacy-safe template updates
• Typical triggers: “install Agent Memory Vault”, “set up memory vault”, “run memory closeout”, “audit my Agent Memory Vault”
• Template repo: mcncarl/agent-memory-vault (https://github.com/mcncarl/agent-memory-vault)

yichen-wechat-mp-batch-exporter

Batch-export WeChat Official Account articles:

• Downloads known mp.weixin.qq.com article URLs as Markdown/JSON/text/HTML
• Uses wechat-article-exporter for account search and history list sync
• Separates publish_groups, expanded_url_items, and original_articles
• Supports enhanced archive planning for read counts, likes, shares, comments, and replies through wxdown-service when fresh user-owned credentials are available
• Requires user confirmation for QR login, credential capture, certificate trust, proxy changes, and any WeChat desktop steps
• Never operates WeChat UI or stores real credentials in the repo

See yichen-wechat-mp-batch-exporter/README.md for setup and privacy notes.

yichen-wecom-local-vault

Read, decrypt, query, and export local WeCom/企业微信 5.x desktop databases on macOS:

• Creates private, timestamped plaintext snapshots and never writes back to the WeCom container
• Supports contacts, sessions, message history, search, and Markdown/JSON export
• Keeps raw keys, snapshots, and chat exports out of Git
• Does not control the original WeCom app or send messages

yichen-grok-consult

Use Grok from a GPT-led Codex task without switching the main model:

• Runs native public X search through the official Grok Build CLI
• Verifies that the isolated Grok session completed XSearch
• Extracts status URLs and deterministically decodes Snowflake publication times
• Keeps Grok outside the current project and disables local file, shell, MCP, memory, and subagent access
• Provides optional independent-answer, review, and challenge tools through local OpenCodex

See plugins/yichen-grok-consult/README.md for installation, privacy boundaries, and verification limits.

yichen-web-research

Top-level router for research tasks that span search, candidate review, archiving, and optional transcription:

• Sends single-stage work directly to the appropriate child Skill
• Never turns a search result into an automatic download
• Adds an evidence-gated horizontal-and-vertical research mode with bounded workstreams, claim-source ledgers, contradiction checks, and retained-gap disclosure
• Routes recent AI discovery, general/vertical web search, platform-native discovery, explicit site maps, and original-source verification through distinct backends instead of silently substituting one for another
• Enforces read-only social-platform use, exact-scope authorization, and no WeChat UI control
• Ships with a portable read-only backend doctor that checks optional adapter readiness without outputting secret values, reading credential-file contents, or making paid probe requests

See yichen-web-research/README.md for the full family, optional backends, and configuration.

yichen-unified-search

Read-only discovery planning across the public web and platform-specific routes:

• Supports AI HOT, AnySearch, GitHub, WeChat public search, Weibo, Xiaohongshu, Douyin, Toutiao, Zhihu, X, Bilibili, YouTube, and Xiaoyuzhou routes
• Uses AI HOT only for time-sensitive AI discovery, AnySearch for general/batch/vertical web search, and Firecrawl only for an explicit bounded site map or explicit verification of a current signed AnySearch candidate
• Provides bounded Zhihu CLI search/hot-list access, anonymous-first public Weibo search, and YouTube keyword/channel discovery through the Data API or a public yt-dlp fallback; it does not download media
• Supports one X Quick call per query and bounded multi-query X Research with phased search, deterministic deduplication, provenance preservation, time-window checks, and at most one gap-filling round
• Built-in adapters produce normalized candidates with provenance, coverage, and limitations; direct GitHub, WeChat Official Account, Xiaohongshu, Douyin, Toutiao, and Bilibili CLI plan steps remain raw unless an explicit downstream normalizer is provided
• Limits browser-session reuse to documented bounded public read-only routes; private-data access and all write actions are outside this Skill

See the detailed platform matrix and route boundaries.

Search query and third-party data flow

Search text is sent to the backend selected for that route. Do not place passwords, cookies, personal data, confidential business information, or private URLs in a search query. When a selected service requires an API or OAuth credential, that credential is sent only to that service according to its protocol; the adapters do not include credential values in candidate output or persist them in