Britneycode/dsh-fish-soul

C3

鱼 · 单角色人格与按人记忆:可纠正、有印象、能导入旧聊天记录,日程自走与主动开口;直连 SnowLuma 收发 QQ(群默认拒绝,要列白名单或开「所有群」才回;没发出去的回复排队重发),B站 / 抖音小号私信与自己刷视频;按功能选模型并统计用量花费,能听会说 QQ 语音、收藏表情包、学群里的梗,每周记忆存档与可选周记,可选上网查。DSH 插件(单角色 feiyu)

★ 10+ · Britneycode/dsh-fish-soul source on GitHub · this plugin in the registry

Britneycode/dsh-fish-soul is a DeepSeek Harness plugin rated C3 — powerful capability combined with sensitive behavior. It patches the dsh runtime, can rewrite the system prompt, command execution in build output only, eval in build output only.

Installable plugin — declares a dsh.bundle manifest

What it can do

CapabilityFlagEvidence
patches the dsh runtimeruntime_patch./cordis.patch.yml
can rewrite the system promptprompt_surfacesystemPrompt, webServer
command execution in build output onlyexec_bundled×5 in build output only, e.g. lib/browsers.js:10, lib/video-frames.js:3
eval in build output onlyeval_bundled×106 in build output only, e.g. lib/douyin-browse.js:219, lib/douyin-browse.js:233
base64 decoding in build output onlybase64_decode_bundled×14 in build output only, e.g. lib/douyin-social.js:384, lib/persona.js:51
reads credential-class env varstoken_envFISH_SOUL_BILI_PUBKEY

Tool registrations

5 static tool registration point(s).

Services it injects

agents llm slots systemPrompt tools webServer workspaceRegistry

Hooks it attaches

agent/inbox/inserted session/event

Outbound domains

www.douyin.com www.bilibili.com message.bilibili.com api.bilibili.com api.siliconflow.cn openspeech.bytedance.com api.github.com passport.bilibili.com creator.douyin.com api.bochaai.com

Environment variables it reads

LOCALAPPDATA FISH_TIME_SCALE DSH_HOST_MODULES APPDATA DSH_HOME FISH_SOUL_SHARP FFMPEG_PATH ProgramFiles FISH_SOUL_BILI_PUBKEY FISH_SOUL_BILI_MOCK

How to read this

Levels measure capability surface and transparency, not maliciousness. A C3 plugin can be entirely legitimate — a desktop shell genuinely needs subprocesses. The point is that you can see this before installing. See the levels explained and how dsh plugins work.

Findings come from static analysis of shipped code; nothing is executed. Think a flag is wrong? Open an issue — every flag cites the file and line it came from.