li-inbox — Claude Code skill: what it does, how to install
li-inbox sorts LinkedIn DMs and connection requests into leads, recruiters, peers, and spam, then drafts replies to send manually.
MIT ★ 3,812 updated 2026-09-17
/plugin marketplace add Jakeschincariol/linkedin-agent-skill
/plugin install linkedin-agent@linkedin-agent-skillWhat li-inbox does
The skill categorizes pasted messages and screenshots from LinkedIn into five distinct buckets: LEAD, RECRUITER, PEER, ASK, and SPAM. It avoids browser automation and never asks to log into your profile. Before proposing actions, it displays exact counts for each group so you can see how much noise arrived.
It flags cold outreach by spotting typical patterns, such as generic invites, calendar links in the first text, or automated follow-ups sent four days later. For real leads and recruiters, it drafts straightforward responses that answer specific questions or request omitted details like salary bands. The user sends every message manually.
What it helps with
- categorize pasted LinkedIn DMs and connection notes
- detect automated outreach sequences and cold pitches
- draft contextual replies for prospective clients and recruiters
- write brief declines for vague requests
Example prompts for the agent
- My LinkedIn inbox is a mess, sort these pasted messages
- Triage my DMs and tell me which ones are automated spam
- Should I reply to this connection request screenshot?
How to install li-inbox
1Plugin marketplace
/plugin marketplace add Jakeschincariol/linkedin-agent-skill/plugin install linkedin-agent@linkedin-agent-skillThis installs the whole linkedin-agent plugin with all its skills
2Skills CLI (Vercel)
npx skills add Jakeschincariol/linkedin-agent-skill --skill li-inbox -a claude-code3Manually: put the files in place
~/.claude/skills/li-inbox/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/li-inbox/
1Skills CLI (Vercel)
npx skills add Jakeschincariol/linkedin-agent-skill --skill li-inbox -a codex2Manually: put the files in place
.agents/skills/li-inbox/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Jakeschincariol/linkedin-agent-skill --skill li-inbox -a cursor2Manually: put the files in place
.cursor/skills/li-inbox/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Jakeschincariol/linkedin-agent-skill skills/li-inbox2Manually: put the files in place
.github/skills/li-inbox/SKILL.md1Skills CLI (Vercel)
npx skills add Jakeschincariol/linkedin-agent-skill --skill li-inbox -a gemini-cli2Platform CLI
gemini skills install https://github.com/Jakeschincariol/linkedin-agent-skill --path skills/li-inboxManually: put the files in place
.devin/skills/li-inbox/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about li-inbox
How to install li-inbox in Claude Code?
Run /plugin marketplace add Jakeschincariol/linkedin-agent-skill followed by /plugin install linkedin-agent@linkedin-agent-skill.
What does the li-inbox skill do?
It splits incoming LinkedIn messages into five categories, detects automated sales sequences, and drafts replies for items worth answering.
Does li-inbox require browser access to LinkedIn?
No, it works strictly on user-pasted text or screenshots and explicitly disallows logging into your account with browser tools.
Source text
SKILL.md · MIT
---
name: li-inbox
description: >-
Triage the LinkedIn inbox - sort connection requests and DMs into leads,
recruiters, peers and spam, and draft the replies worth sending. Use when the
user says "my inbox is a mess", "triage my DMs", "should I reply to this",
pastes a batch of LinkedIn messages, or is drowning in connection requests.
---
# li-inbox
Most LinkedIn inboxes are 80% noise, and the cost of that noise is that the
20% goes unanswered for a week. This skill separates them, then writes only
what is worth writing.
## Input
The user pastes the messages. Screenshots are fine. Do not log into their
account or read their inbox with a browser tool.
## Sort into five
| bucket | signal | action |
| --- | --- | --- |
| **LEAD** | describes a problem the user solves, or asks about working together | reply today, full answer |
| **RECRUITER** | a role, a company, a salary band | reply if the role is real, one line if not |
| **PEER** | someone in the same field with something to say | reply this week, keep it human |
| **ASK** | wants advice, time, an intro, a favour | reply if it is cheap and specific, decline cleanly if not |
| **SPAM** | agency pitch, lead-gen sequence, crypto, "quick question" with no question | archive, no reply |
Print the counts first. Seeing "3 leads, 2 recruiters, 41 spam" is most of the
value.
## Detecting a sequence
Automated outreach has a shape: an invite note with no specifics, a message
that arrives within minutes of the accept, "quick question", "I noticed you're
in {industry}", a calendar link in message one, then a bump exactly four days
later. When you see it, mark it SPAM and say which tell gave it away. The user
does not owe a reply to a script.
## Replies
- **LEAD** - answer the actual question in the message, in full, for free. If
it is a fit, the offer is one sentence at the end. If it is not, say so and
point them somewhere useful. Both outcomes are good.
- **RECRUITER** - if the role is genuinely interesting, ask the three things
the message left out: comp band, level, and whether it is in-office. If it
is not, one line: not looking, happy to refer, and mean the refer.
- **ASK** - if it costs under ten minutes and is specific, do it. If it is
"can I pick your brain", decline in one warm sentence and give them the one
answer you would have given on the call. That is the polite version and it
is also the more useful one.
- **DECLINES** are short, warm and final. No "let's revisit in Q3" if there is
no Q3.
## Output
Grouped by bucket, counts first, drafts only for the buckets that get replies,
each one humanized. Then the gate: the user sends them.
```
INBOX · 52 items · 3 LEAD, 2 RECRUITER, 4 PEER, 2 ASK, 41 SPAM
SPAM (41) - archive. 38 are the same sequence: no-specifics invite,
"quick question" within 4 minutes of accept, calendar link in message one.
```
© Jake Schincariol · MIT · Jakeschincariol/linkedin-agent-skill