Jianshuo Wang (王建硕)

claude-skills

Audio/video → timestamped SRT in the source language. Chinese uses Volcano ASR; other languages use OpenAI Whisper with word-level timestamps + punctuation-bounded re-segmentation.

Translate an SRT from one language to another with punctuation-bounded re-segmentation. Simplified Chinese and English are first-class targets. Supports bilingual output.

Time-aligned TTS dub of a video against a target-language SRT. Volcano TTS for Chinese, edge-tts for everything else. Single-speaker by default; opt-in visual diarization for multi-speaker.

Final-composite step: burn SRT into pixels via libass, mix in a dub track, keep the original audio as a low-volume bed — all in one ffmpeg encode (no cascade).

Thin orchestrator for the end-to-end video localization pipeline (transcribe → translate → dub → burn). Use it when the user asks for full localization in one go.

Audio cross-correlation alignment for N recordings of the same event. Outputs `.sync.json` sidecars per file — never re-encodes the originals.

Auto-edit 2+ synced cams into a single MP4. Switches between angles second-by-second based on audio energy. Optional picture-in-picture.

Cut a long-form interview / lecture / podcast into 3–6 stand-alone topical short clips, with per-clip SRT — ready for downstream post-production.

Post-production overlays on top of a clip: AI-generated cover as first frame, HTML/CSS kinetic captions, illustration overlays, chapter chips, end-card CTA. HyperFrames-backed, single final encode.

Horizontal ↔ vertical conversion (16:9 ↔ 9:16, 4:3 ↔ 3:4). Crops a narrow band and tracks the active speaker via MediaPipe + mouth-aspect-ratio variance.

Batch upload videos to YouTube with per-video metadata from `UPLOAD_META.md`. Survives SOCKS/HTTP proxies by using `requests` directly for resumable upload.

Turn a 王建硕-style WeChat article into a narrated 1080×1920 portrait MP4 (30–90s) — Volcano TTS voiceover, HyperFrames CSS/GSAP per-scene animation, subtle SFX, abstract watercolor background.

Turn a single English word into an HLS supercut lesson from a video base — stitches every clip where the word is spoken, prepended with a bilingual intro card (word + IPA + 中文 gloss + usage, Volcano TTS) and a follow-me CTA card.

Polish a 微信公众号 (WeChat Official Account) article, generate a cover image (题图) and explanation illustration, and upload as a draft to mp.weixin.qq.com.

Mine a video's SRT — a monologue / 讲解, or an interview where 王建硕 is one speaker — into multiple standalone 微信公众号 articles, one per distinct topic.

Turn uploaded VoiceDrop voice memos (the jianshuo.dev/files R2 inbox) into WeChat article drafts — pull the unprocessed recordings, transcribe, and mine articles from each.

Capture a writer's voice from a few sample articles and rewrite text to sound like that author. Works for any author, 王建硕 included.

Auto-syndicate your newest 微信公众号 article across platforms — auto-posts to X / Bluesky / Threads / LinkedIn and prepares a copy-paste outbox for manual platforms (Facebook / 小红书 / 即刻 / 知乎).

Post a daily X tweet inspired by your newest un-tweeted WeChat article — drafts 3 angle variants (quote / metaphor / one-liner), posts the chosen one via xurl, records history.

Systematically improve X content by A/B-testing the tweet-generation prompt — git-SHA-versioned, numbered experiments judged on median impressions per tweet, with per-tweet feature mining.

Grow X followers via numbered, A/B-testable growth experiments — each with a hypothesis, target metric, rollback state, and verdict. North-star = new-follower ÷ profile-visit conversion.

Rewrite a plain Chinese tweet into a high-engagement version backed by real facts + an interaction hook. Returns 2–3 variants, short / plain / colloquial.

Delete 引流 / emoji spam replies (同城面付 / 寻固炮 / 线下上门 etc.) under your X posts. Covers the last 7 days via X recent-search.

Conversational backend for a Hugo static blog — add / edit posts, manage categories, upload images, commit, push, auto-deploy. No CMS, no server.

Migrate a WordPress site (WXR export + uploads) to a Hugo static site on GitHub Pages — preserves /archives/<id>/ URLs, localizes images, deploys via GitHub Actions.

Wire up fastlane + GitHub Actions for an iOS project so pushing main auto-builds and uploads to TestFlight.

Ship a TestFlight iOS app to the App Store — prepare screenshots + description/metadata and submit for review via the existing fastlane setup.

Add an in-site feedback loop to a website repo — a floating 提个建议 button where allowlisted visitors' suggestions become GitHub Issues that Claude Code turns into auto-merged, auto-deployed code changes, with a /_feedback dashboard + one-click revert.

Holistic state-of-the-project audit: unmerged branches, stalled PRs, failed GitHub Actions, stale builds, plan drift, unreleased commits, log errors. Read-only investigation → grouped checklist → fix only after confirmation.

Automated daily promotion for your Claude Code skills — picks a skill, generates a marketing angle, auto-posts to X via xurl, and drafts community posts (Reddit / HN / Discord) to an outbox.

吃一堑长一智 — 6-step interactive reflection for personal setbacks. Diagnoses which layer (L1 notes / L2 in-the-moment trigger / L3 default-reaction practice) is missing and routes to the right modality, instead of letting every reflection become another notebook entry.

VoiceDrop 配套 skill。两个模式 — (1) list：列出你在 VoiceDrop R2 上的全部文件（录音 / 文章 JSON / 标记）；(2) distill：从你的样本文章蒸馏文风，生成可贴入 VoiceDrop「设置→文风」的精简规则集，并可直接通过 Files API 上传。触发词："voicedrop list"、"列出 VoiceDrop 文件"、"voicedrop distill"、"蒸馏 VoiceDrop 文风"、"/wjs-voicedrop"。

wjs-evaling-voicedrop-prompts