// linkedin + x · one CLI for agents
Give your agent access to social networks.
Copy one setup prompt. Your agent installs the CLI, starts device login, and gets LinkedIn and X ready from the shell it already uses.
no keys to manage · one bill · auditable
// capabilities
Hand your agent the things you keep meaning to do.
Ask in plain language. Your agent runs the calls — reading and posting through your own account, on your terms.
// graph
Explore your network.
ask your agent
LinkedIn · X
Explore my LinkedIn connections and my newest X followers, then surface the ten people worth reaching out to this week.
// reactions
See who's in the room.
ask your agent
Pull everyone who reacted to my last LinkedIn post and tell me which ones look like warm leads.
// study
Learn from people who post better.
ask your agent
LinkedIn · X
Read this creator's recent posts and break down what makes them land, so I can write sharper ones.
// inbox
Clear your LinkedIn inbox.
ask your agent
Go through my LinkedIn inbox, prioritize the messages that matter, and draft replies I can send myself.
// dms
Stay on top of X DMs.
ask your agent
X
Read my unread X DMs, group the ones that need attention, and draft responses in my voice.
// listen
Listen across both networks.
ask your agent
LinkedIn · X
Find what people are saying about AI agents on LinkedIn and X this week and summarize the themes for me.
// example · one shell pipe
A real agent task, in shell you can read.
warm-contacts.sh
# Find hiring posts, fan out to reaction graphs,
# emit a ranked list of warm contacts.
social linkedin search posts "AI infra hiring" --limit 10 --json \
| jq -r '.posts[].urn' \
| xargs -I{} social linkedin posts reactions {} --json \
| jq -s '
[.[].reactions[] | { actor: .actor, type: .type }]
| group_by(.actor.id)
| map({ actor: .[0].actor, signals: length })
| sort_by(-.signals)
| .[0:15]
'Composable with the tools you already use.
Every command speaks JSON. Pipe through jq, fzf, your agent's planner — anything that reads stdin.
Same flags on LinkedIn and X.
--limit, --account, --json, --no-cache. Learn the surface once; reuse it across networks.
Designed for agents first.
Deterministic exit codes, structured errors, and a SKILL.md you drop straight into Claude, Codex, or any other agent.
// account safety
Built so your accounts outlive your prototype.
The unspoken question: will it kill my LinkedIn? Here's the short answer.
// scope
Read scope by default.
Writes need an explicit upgrade you can revoke. A read-only session can't post, invite, or message.
// writes
A write never fires twice.
LinkedIn's 429s and 500s surface verbatim and we never retry a write — so an invite or DM can't double-send.
// proxy
Residential proxy, per account.
Every account runs over a dedicated residential IP with smart rate-limiting, so the platform sees a normal human session — not a datacenter.
// agent setup prompt
Give your agent the setup prompt.
Copy this into Claude Code, Codex, or any terminal agent. It installs the CLI, starts login, and leaves browser approval with you.
agent-setup.prompt
Install the social CLI for this workspace.Use Bun to install the CLI globally:bun install -g @usesocial/cliThen run:social loginWhen the CLI prints a device approval URL like https://usesocial.dev/device?code=K3R-7QX or a code, give it to me so I can approve the session in my browser.After login succeeds, load the bundled social agent skills for LinkedIn and X so you can use the CLI safely from this workspace. Prefer JSON output for commands you run, and ask me before requesting write scope or connecting a new social account.