dsh-linkedin-agent
适用于 DeepSeek Harness 的 11 个 LinkedIn 代理技能（Jakeschincariol/linkedin-agent-skill 的移植版），MIT。
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:mubaid/dsh-linkedin-agent#d11b90454ce0a3350686b01a021190abe8565f25Linkedin Agent 以 dsh-linkedin-agent 发布，当前版本为 0.1.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Eleven LinkedIn agent skills in DeepSeek Harness — posts off 21 hook formulas, comments, replies, a 100-point profile score, a weekly plan, and a humanizer that strips the AI fingerprint before anything goes out. Free, MIT, no signup.
English | 简体中文
You want to run a LinkedIn presence with an agent. You write posts, comment on other people's threads, reply under your own, score your profile, and plan the week. That is eleven workflows, and doing them well takes consistent voice and judgement. Doing them sloppily makes you look like a template.
This plugin installs eleven li-* skills. They write everything — you post
everything. Nothing reaches LinkedIn without your thumbs-on-keyboard yes.
dsh plugin add github:mubaid/dsh-linkedin-agent
Restart the profile. /li-post is ready.
/li-post I shipped a proposal template that cut prep time in half
HOOKS
1. #17 Time Anchor Writing a proposal used to take me half a day. It now takes 20 minutes.
2. #12 Comparison A freelance consultant vs a weekend and a template. The weekend won.
3. #3 Mistake For two years I charged for hours I spent fixing my own messy process.
Post #17. It is a ratio, and the number is yours.
Three hook options, one draft on the strongest, humanized before you see it.
The humanizer is the whole point. Every post, comment, reply and DM runs
through /li-human first — em dashes to commas, the 113-word AI slop lexicon,
zero-width fingerprint characters — and scores on a five-check panel. A
post that passes the panel is readable; a post that fails comes back for
human work, not a second polish.
No automation theatre. These skills do not post to LinkedIn, and they should not: there is no approved API for posting to a personal profile, and browser automation violates LinkedIn's User Agreement and gets accounts restricted. Every skill ends with a copy-ready block. You paste. That is the design, not a bolted-on limit.
Nothing fabricated. If a draft needs a number you haven't given, it comes
back with {{your number}} and a flag, every time. No invented metrics,
clients or outcomes.
Voice, actually. /li-plan and the rest read templates/voice.md — fill
it in first, or paste three of your own posts and say "write my voice.md from
these". Every skill reads it. Skip it and everything comes out generic.
dsh plugin add github:mubaid/dsh-linkedin-agent
dsh --profile <your-profile> agent
/li-post <an idea>
Ten minutes on templates/voice.md first.
Writing posts that sound like you. Fill in the voice file with three of your own posts, and the gap between "drafts I post" and "drafts I rewrite" widens immediately.
Engagement rounds. /li-plan builds a 10-person engage list (5 reach, 3
peers, 2 buyers); /li-comment does the commenting; /li-inbox keeps the
pipeline honest.
Before you post, prove it's human. /li-human on the final draft:
BURSTINESS, SPECIFICITY, SLOP DENSITY, FINGERPRINT, VOICE — the
verdict weights the mean at 60% and the weakest check at 40%.
The skills are registered through dsh-skill-filesystem under provider name
linkedin-agent, scanning the skills/ directory next to lib/. Each skill is a SKILL.md
reproduced byte-for-byte from upstream.
| Claude Code install | Proxy / aggregator | This plugin | |
|---|---|---|---|
| Harness | Claude Code only | Any | DeepSeek Harness |
| Install surface | ~/.claude/skills/ | Any | dsh plugin add |
| Humanizer built in | Yes | Depends | Yes (local, runs on your machine) |
| Posts to LinkedIn | No | Depends | No |
| Upfront cost | Clone + copy | Key + route | Zero config |
templates/voice.md — three of your own posts, or a description of
your style. It is the single biggest factor in whether the output sounds
like you.UPSTREAM.md for the full file-by-file mapping to the upstream
Jakeschincariol/linkedin-agent-skill.0.2.0-rc.2 and what is still pending.MIT. See LICENSE and NOTICE. Port of Jakeschincariol/linkedin-agent-skill.