🐳 dsh-whale

Big whale, few words.

Minimal speech mode plugin for DeepSeek Harness (DSH) — cut all AI fluff, save 60-75% output tokens

📊 Before / After

Normal (69 tokens)	Whale (18 tokens)
"Sure! I'd be happy to help. The reason your React component re-renders is likely because you're creating a new object reference on each render. I recommend wrapping it in useMemo."	"New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo."

74% fewer tokens, zero information loss.

# From GitHub (recommended)
dsh plugin --profile web add github:1Lyn-en/dsh-whale

# From npm
dsh plugin --profile web add @1lyn-en/dsh-whale

Then restart dsh web and type /whale in chat to activate.

Command	Mode	Effect	Savings
`/whale lite`	🐋 Lite	Cut pleasantries, keep full sentences	~40%
`/whale` or `/whale full`	🐳 Full (default)	Drop articles, fragment-style	~60%
`/whale ultra`	🦈 Ultra	One word when one word enough	~75%
`/whale wenyan-lite`	📜 Classical Lite	Semi-classical Chinese	~60%
`/whale wenyan-full`	🏮 Classical Full	Pure classical Chinese	~85% (chars)
`/whale off`	🌊 Off	Back to normal	—
`/whale status`	—	Show current mode	—

Automatically exits whale mode temporarily for:

Safety > tokens.

Also available as a SKILL.md for Claude Code / Codex / Cursor:

/skill add https://raw.githubusercontent.com/1Lyn-en/dsh-whale/master/cross-platform/SKILL.md

npm install
npm run typecheck
npm test
npm run build
npm run pack

🐳 Big whale, few words.