LinkedIn Lead-Alerts → 3 Comment Drafts → Slack

Daily pipeline that drafts 3 comment variants (in Petro's voice via the linkedin-comment-ideas skill) for each new post by a Sales Navigator lead, and posts them to Slack for review/redaction before you paste back to LinkedIn.

What it does

1. daily_run.py (launchd, 09:00 Mac Studio) scrapes the Sales Navigator lead-alerts feed with Playwright using a saved logged-in session.
2. For each new post, shells out to the claude CLI with the linkedin-comment-ideas skill to generate 3 variants.
3. Posts each post + 3 variants into #linkedin-replies (top-level = post; thread = variants).
4. You react 1️⃣ / 2️⃣ / 3️⃣ to pick. Reply in thread with edit notes.
5. A scheduled remote agent (every 15 min) applies edits and posts *FINAL —* in the thread. You copy/paste into LinkedIn yourself.

One-time setup (Mac Studio)

# 1. Python deps
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
playwright install chromium

# 2. Skill (linkedin-comment-ideas) — symlink so `git pull` refreshes it
git clone git@github.com:boarlabsxyz/linkedin-ai.git ~/repos/linkedin-ai
ln -s ~/repos/linkedin-ai/.claude/skills/linkedin-comment-ideas \
      ~/.claude/skills/linkedin-comment-ideas

# 3. Configure MCP servers Claude Code needs for the skill
#    (Google Drive + Google Docs — for the Posted/Transcripts folders the
#    skill references). Use `claude mcp add` per your auth flow.

# 4. Environment
cp .env.example .env
# Fill in SLACK_BOT_TOKEN, SLACK_CHANNEL_ID, SLACK_HUMAN_USER_ID.

# 5. LinkedIn login — opens a browser; log in (including 2FA); press Enter.
python linkedin_login.py
chmod 600 storage_state.json

# 6. Smoke test (no Slack writes)
python daily_run.py --dry-run

# 7. Install the launchd job
cp com.user.linkedin-replies.plist ~/Library/LaunchAgents/
launchctl bootstrap gui/$UID ~/Library/LaunchAgents/com.user.linkedin-replies.plist

# 8. Register the remote Slack-poll routine — paste the prompt from
#    remote_routine.md into `/schedule` (cron: */15 * * * *).

Slack bot setup

Create a Slack app (https://api.slack.com/apps), add bot token scopes:

• chat:write
• channels:history (or groups:history if #linkedin-replies is private)
• channels:read
• reactions:read

Install to workspace, copy the Bot User OAuth Token (xoxb-…) into .env, invite the bot to #linkedin-replies.

Files

File	Purpose
linkedin_login.py	One-time headed browser to capture storage_state.json.
daily_run.py	Daily scrape + generate + post. Supports --dry-run.
com.user.linkedin-replies.plist	launchd schedule.
remote_routine.md	Prompt for the scheduled remote agent that finalizes replies.
state/seen_posts.json	Dedup ledger (auto-created).
storage_state.json	Playwright session cookies. Sensitive — chmod 600.
logs/	Daily logs.

Re-auth

LinkedIn cookies expire (~30 days). If daily_run.py detects a login-wall it posts a heads-up in Slack — re-run python linkedin_login.py to refresh.