LinkedIn Auto-Reply Bot

Cross-platform CLI that auto-replies to new comments on your own LinkedIn posts and, optionally, sends a follow-up DM to the commenter. Ships as a linkedin-autoreply command with a guided init wizard.

---

⚠ ToS Disclaimer

LinkedIn's User Agreement prohibits automated access and interactions with the platform. This tool operates in a gray zone — it acts only on your own posts, at low volume, from your own device and IP. Risk is not zero.

By running this tool you accept that LinkedIn may restrict, suspend, or permanently ban your account. Use at your own risk.

---

Platform Support

Platform	Status
macOS (Apple Silicon + Intel)	✅ Developed & live-tested
Windows 10/11	⚠️ Code ready, live validation pending — unit tests pass via subprocess mocks, but no end-to-end run on a real Windows host yet. Please open an issue if something breaks.
Linux	⚠️ CLI + bot work, but no scheduler integration (launchd is macOS-only, Task Scheduler is Windows-only). You need to run linkedin-autoreply run from cron/systemd yourself.

---

Requirements

• Python 3.11 or newer
• Active LinkedIn account (you'll log in once, session cookies are stored locally)
• ~500 MB free disk space (Chromium browser for scrapling + virtualenv)

---

One-Line Install

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/JSap0914/linkedin-automation/main/install.sh | bash

Windows (PowerShell)

iex (iwr -useb https://raw.githubusercontent.com/JSap0914/linkedin-automation/main/install.ps1).Content

What the installer does:

• clones to ~/.linkedin-automation
• creates .venv
• runs pip install -e ".[dev]"
• runs scrapling install
• creates a global wrapper command (linkedin-autoreply) that always runs from the install root
• immediately launches linkedin-autoreply init

If the install directory already exists as a clean git checkout, the installer fast-forwards it. If it has uncommitted changes, it aborts rather than overwriting them.

---

Manual Install

Any Python 3.11+ works. You do not need to "activate" the venv — just call its binaries directly. This matches what install.sh / install.ps1 do internally and sidesteps PowerShell's Activate.ps1 pitfalls entirely.

macOS / Linux

git clone https://github.com/JSap0914/linkedin-automation.git
cd linkedin-automation
python3 -m venv .venv
.venv/bin/python -m pip install -e ".[dev]"
.venv/bin/scrapling install
.venv/bin/linkedin-autoreply init

Windows (PowerShell)

git clone https://github.com/JSap0914/linkedin-automation.git
cd linkedin-automation
py -3 -m venv .venv
.venv\Scripts\python.exe -m pip install -e ".[dev]"
.venv\Scripts\scrapling.exe install
.venv\Scripts\linkedin-autoreply.exe init

If you really want to activate the venv in PowerShell, you must prefix the path with .\ and allow local scripts:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy RemoteSigned
.\.venv\Scripts\Activate.ps1

Without .\, PowerShell treats .venv as a module name and fails with CouldNotAutoLoadModule.

---

What the init Wizard Does

Everything is pre-configured with sensible defaults. You only answer 2 questions — the rest is automatic.

1. ToS acknowledgement — accept the automation risk (must type y)
2. Prerequisite check — (automatic) Python version + scrapling importable
3. LinkedIn login — opens a Chromium window; log in manually (you have 5 minutes); session cookies are cached in .profile/
4. Enable auto-DM? [Y/n] — sends a follow-up DM to commenters (1st-degree only, max 30/day)
5. Write config — (automatic) saves replies.yaml with defaults
6. Bootstrap — (automatic) marks all existing comments as "seen" so the bot won't reply to old ones
7. Install scheduler — (automatic) registers a launchd agent (macOS) or Task Scheduler task (Windows) that runs the bot every 60 seconds. On Linux, prints a notice to set up cron/systemd manually.
8. Done — prints commands for customizing settings

Defaults written to replies.yaml

Setting	Default	Change later with
Auto-reply enabled	true	config set enabled false
Reply templates	3 Korean thank-you messages	config wizard or config edit
Reply delay	0 s (instant)	config set reply_delay_seconds_min 30
Polling interval	60 s	config set polling_min_interval_seconds 300
Post lookback	30 days	config set post_lookback_days 7
DM: 1st-degree only	true	config set dm.only_first_degree_connections false
DM: auto-accept invites	true	config set dm.auto_accept_pending_invitations false
DM: max per day	30	config set dm.max_per_day 50
DM: delay	0 s	config set dm.delay_seconds_min 60

To re-configure interactively, run linkedin-autoreply config wizard anytime.

What gets created on disk

Path	What it is	Safe to delete?
replies.yaml	Your configuration	No (regenerate via config reset)
.profile/	Chromium profile + cookies (~100 MB)	No (you'll need to re-login)
.cache/own_urn	Cached LinkedIn person URN	Yes (auto-regenerated)
seen_comments.db	SQLite log of replied comments	Yes (but you may re-reply to old comments)
logs/bot.log	Rotating bot log	Yes
~/Library/LaunchAgents/com.user.linkedin-autoreply.plist (macOS)	Scheduler entry	Yes (run uninstall instead)
Task Scheduler LinkedInAutoReply (Windows)	Scheduler entry	Yes (run uninstall instead)

---

All Commands

linkedin-autoreply init              # full setup wizard
linkedin-autoreply run [--dry-run] [--bootstrap]
linkedin-autoreply setup             # re-login only (no config changes)
linkedin-autoreply update [--dry-run] [--skip-tests]

linkedin-autoreply config show
linkedin-autoreply config set <dotted.path> <value>
linkedin-autoreply config edit       # opens in $EDITOR
linkedin-autoreply config wizard     # re-run config section only
linkedin-autoreply config reset      # overwrite with defaults
linkedin-autoreply config migrate    # fill missing fields after schema change

linkedin-autoreply start             # install + enable scheduler
linkedin-autoreply stop              # disable (keep entry)
linkedin-autoreply uninstall         # remove scheduler entry
linkedin-autoreply status            # scheduler state + recent logs
linkedin-autoreply logs [-n N]       # tail bot.log

Config paths you'll actually change

linkedin-autoreply config set dm.enabled true
linkedin-autoreply config set dm.max_per_day 50
linkedin-autoreply config set enabled false              # kill switch
linkedin-autoreply config set reply_delay_seconds_min 30
linkedin-autoreply config set reply_delay_seconds_max 120
linkedin-autoreply config set polling_min_interval_seconds 300  # 5 min

---

Verifying Your Install Works

Run these in order after init:

# 1. Config is valid
linkedin-autoreply config show

# 2. Scheduler is registered
linkedin-autoreply status
# → should show "Installed: yes" + "Enabled: yes"

# 3. Bot can actually call LinkedIn (no write, safe)
linkedin-autoreply run --dry-run
# → logs should show "Fetched N comments" and "Dry run complete"

# 4. Check logs
linkedin-autoreply logs -n 20

# 5. (macOS) verify launchd picked it up
launchctl list | grep linkedin-autoreply
# → should print one line

# 6. (Windows PowerShell) verify Task Scheduler
schtasks /Query /TN LinkedInAutoReply

---

Updating to the Latest Version

linkedin-autoreply update

If you installed via install.sh / install.ps1, this works from any current working directory because the global wrapper command always cds into the install root before invoking the CLI.

This:

1. Aborts if your working tree is dirty (run git status to see)
2. git pull --ff-only origin main (refuses non-fast-forward)
3. Re-runs pip install -e ".[dev]" if pyproject.toml changed
4. Runs config migrate if bot/config.py or bot/config_defaults.py changed (fills new fields with defaults, drops removed fields — your values are preserved)
5. Reinstalls the scheduler if bot/scheduler/templates/ changed
6. Runs the pytest smoke suite (--skip-tests to skip)

Flags:

• --dry-run — inspect only, no pull/install/migrate
• --skip-tests — don't run pytest after update

---

Kill Switch

Stops the bot without uninstalling:

linkedin-autoreply config set enabled false

The next scheduled run exits cleanly with no LinkedIn API calls. Re-enable with ... set enabled true.

---

Full Uninstall

linkedin-autoreply uninstall   # removes scheduler entry
deactivate                     # exit venv
rm -rf .venv .profile .cache seen_comments.db logs replies.yaml
cd ..
rm -rf linkedin-automation     # manual-install path

If you used the one-line installer, remove the installer-managed directory instead:

linkedin-autoreply uninstall
rm -rf ~/.linkedin-automation
rm -f ~/.local/bin/linkedin-autoreply

Windows (PowerShell):

linkedin-autoreply uninstall
Remove-Item -Recurse -Force $HOME\.linkedin-automation
Remove-Item -Force "$HOME\AppData\Local\Microsoft\WindowsApps\linkedin-autoreply.cmd" -ErrorAction SilentlyContinue

---

Migration from Pre-CLI Workflow

Before	After
python setup.py	linkedin-autoreply setup (setup.py is gone)
bash launchd/install.sh	linkedin-autoreply start (bash script is a shim)
bash launchd/uninstall.sh	linkedin-autoreply uninstall (shim)
edit replies.yaml by hand	linkedin-autoreply config set … or config edit
manual re-bootstrap	linkedin-autoreply config reset then run --bootstrap

python bot.py [--dry-run] [--bootstrap] still works as a compatibility shim.

---

Troubleshooting

scrapling install fails / browser not found

scrapling install --force   # re-download Chromium

If behind a corporate proxy, set HTTPS_PROXY first.

AuthExpiredError in logs

Session cookie expired (LinkedIn rotates them every ~1 year, or sooner if you log in elsewhere):

linkedin-autoreply setup

Login step hangs

Browser window didn't open? Check that you have a display server. Over SSH you need -X forwarding or a local terminal.

"Another instance holds logs/bot.lock"

A previous run crashed. Safe to delete:

rm logs/bot.lock logs/bot.lock.lock 2>/dev/null

Bot replies to comments from weeks ago

You skipped bootstrap. Run:

linkedin-autoreply run --bootstrap

This marks every current comment as "seen" without replying.

linkedin-autoreply: command not found

Call the binary inside the venv directly — no activation needed.

macOS / Linux:

.venv/bin/linkedin-autoreply --help
.venv/bin/python -m pip install -e ".[dev]"   # if deps missing

Windows (PowerShell or cmd):

.venv\Scripts\linkedin-autoreply.exe --help
.venv\Scripts\python.exe -m pip install -e ".[dev]"

PowerShell: '.venv' 모듈을 로드할 수 없습니다 / CouldNotAutoLoadModule

You ran .venv\Scripts\Activate.ps1 without the leading .\. PowerShell treats .venv as a module name instead of a relative path. Two fixes:

Easiest — skip activation entirely (recommended; matches what our installer does):

.venv\Scripts\python.exe -m pip install -e ".[dev]"
.venv\Scripts\scrapling.exe install
.venv\Scripts\linkedin-autoreply.exe init

If you insist on activating, prefix with .\ and allow local scripts:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy RemoteSigned
.\.venv\Scripts\Activate.ps1

-Scope Process only lasts for the current terminal session.

PowerShell: cannot be loaded because running scripts is disabled on this system

Your execution policy is Restricted (Windows default). Either:

# Temporary — only this terminal session
Set-ExecutionPolicy -Scope Process -ExecutionPolicy RemoteSigned

# Permanent for your user — no admin needed
Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned

Or just use the venv binaries directly (no activation, no policy change needed).

Python 3.11+ is required / No suitable Python runtime found (Windows)

The installer prefers python (then python3, explicit versions, then py -3.X). Microsoft Store python stubs under WindowsApps\ are deliberately skipped because they open the Store instead of running Python.

"No suitable Python runtime found" is emitted by py.exe itself when its registry (HKCU/HKLM\SOFTWARE\Python\PythonCore) doesn't list the requested version — even if you actually have Python installed. Common causes:

1. You installed via the Microsoft Store — py.exe doesn't see Store installs by default.
2. You installed from python.org but unchecked "Install py launcher" or "Add Python to PATH".

Diagnose what py sees:

py -0
$env:PYLAUNCH_DEBUG = 1; py -3.13 -c "import sys"

Fix (one of):

• Re-run the python.org installer with both "Add python.exe to PATH" and "Install py launcher" checked.
• Or just run the install script with python directly instead of relying on py:

  git clone https://github.com/JSap0914/linkedin-automation.git $HOME\.linkedin-automation
  cd $HOME\.linkedin-automation
  python -m venv .venv
  .venv\Scripts\python.exe -m pip install -e ".[dev]"
  .venv\Scripts\scrapling.exe install
  .venv\Scripts\linkedin-autoreply.exe init

On macOS/Linux, check what you have:

command -v python3.11 python3.12 python3.13 python3

Install a recent CPython from python.org if none show up.

Tests fail with ModuleNotFoundError: filelock

pip install -e ".[dev]"  # reinstall deps

launchd (macOS) won't start

launchctl list | grep linkedin-autoreply
# If present but "Status: 78" or similar error code:
launchctl unload ~/Library/LaunchAgents/com.user.linkedin-autoreply.plist
linkedin-autoreply start  # reinstalls

Windows Task Scheduler blocks the task

schtasks /Create requires the user to have "Log on as a batch job" right. Some corporate Windows images disable this. Workaround: run linkedin-autoreply run manually from an elevated PowerShell, or skip the start step and invoke run via your own scheduler.

---

How It Works

Uses LinkedIn's internal Voyager API (/voyager/api/voyagerSocialDashNormComments for replies, /voyager/api/voyagerMessagingDashMessengerMessages for DMs) — the same API the web app calls. Authenticates via browser session cookies stored in .profile/ (not username/password).

Every 60 seconds the OS scheduler runs one poll cycle:

1. Load session cookies from .profile/
2. Fetch your posts from the last N days (default 30)
3. For each post, fetch comments
4. Filter out: your own comments, nested replies, already-seen comments (tracked in seen_comments.db)
5. For each new comment:
   • Pick a reply template (per-post binding > keyword match > random default)
   • Personalize with {name}님 (Korean honorific) using the commenter's first name
   • Wait reply_delay_seconds (0 = instant)
   • POST the reply
   • If dm.enabled + commenter passes only_first_degree_connections check: auto-accept any pending invitation from them, wait for connection, then send a DM
   • Record success in seen_comments.db

All pre-existing comments at the time of linkedin-autoreply init (bootstrap step) or run --bootstrap are marked seen without reply. New comments landing after that moment are what the bot responds to.

See Scrapling for the underlying Patchright-based browser automation.

---

Known Limitations

• Single LinkedIn account per install — multi-account would need separate clones
• No retry on Voyager 429 — the bot logs + exits; next scheduled run picks up where it left off
• No rich text replies — plain text only (LinkedIn supports mentions but this bot doesn't emit them)
• Korean-first templates — defaults are in Korean ({name}님 ...); override via wizard or config set
• Windows live validation pending — see Platform Support above

---

Project Layout

bot/
├── cli.py                    # Typer CLI entry point
├── cli_commands/             # One file per subcommand
├── onboarding/               # init wizard steps
├── scheduler/                # launchd / Task Scheduler abstraction
│   └── templates/
├── config.py                 # pydantic schema
├── config_defaults.py        # single source of truth for defaults
├── config_io.py              # YAML read/write + dotted-path setter
├── config_migrate.py         # schema drift detection + safe merge
├── updater.py                # git pull + pip install + drift detection
├── orchestrator.py           # main poll cycle
├── auth.py                   # browser login + cookie extraction
├── voyager.py                # Voyager API client (scrapling)
├── comments.py, posts.py, replies.py   # endpoint wrappers
├── messaging.py, connections.py, invitations.py  # DM + connection logic
├── templates.py, personalization.py    # template matching + {name} substitution
├── db.py                     # SQLite seen_comments / dm_sent
├── lockfile.py               # cross-platform singleton lock (filelock)
├── killswitch.py
└── logging_config.py

tests/                        # 255 unit tests (pytest)
launchd/                      # macOS launchd shim scripts (legacy path)
pyproject.toml                # deps, entry points, pyright config

---

Contributing / Issues

Found a bug, especially on Windows? Open an issue. Include:

• OS + Python version
• Output of linkedin-autoreply status
• Last 50 lines of logs/bot.log
• What you ran + what you expected

---

License

MIT