By name: linkedin-sourcing description: LinkedIn Recruiter (paid product) sourcing assistant for macOS. Excel-driven state, phased execution, resumable workflows, fast chained browser automation. license: MIT metadata: author: martinliu version: "1.2.0" allowed-tools: Bash(npx agent-browser:), Bash(agent-browser:), Bash(grep:), Bash(rg:), Bash(ls:), Bash(mkdir:), Bash(cat:), Bash(echo:), Bash(date:), Bash(timeout:), Bash(python3:), Bash(open:), Bash(chmod:), Bash(bash:)
LinkedIn Sourcing
Golden Rule: Run the Loop
For normal operation, always run:
python3 "$SKILL_DIR/scripts/run_reachout_loop.py" --project "{PROJECT_ID}"
The loop decides the next phase from project and workbook state. Do not run individual phase scripts to advance a project.
If you are trying to make progress, run the loop. If you are running anything else, you are debugging a loop failure.
Quick Reference
Bootstrap a new project
python3 "$SKILL_DIR/scripts/bootstrap_project.py" --jd-url "https://example.com/job"
Agent responsibility during bootstrap:
- Read the job description (JD) from the provided URL
- Infer
CORE_FUNCTION(what the team does, e.g., "building scalable ML infrastructure") from the JD - Infer
BUSINESS_IMPACT(why it matters, e.g., "powering recommendation systems for billions of users") from the JD - Pass them as CLI flags:
python3 "$SKILL_DIR/scripts/bootstrap_project.py" --jd-url "..." --core-function "..." --business-impact "..." - Only ask the user for these fields if the JD is ambiguous or you are genuinely uncertain after reading it
Start or continue work
python3 "$SKILL_DIR/scripts/run_reachout_loop.py" --project "{PROJECT_ID}"
After any stop
- Read the loop's stop reason.
- Fix only that blocker.
- Run the same loop command again.
Do not guess the next phase.
Stop Handling
| Stop reason | What to do |
|---|---|
| Login/auth required | Log into LinkedIn Recruiter, then rerun the loop |
| Browser/CDP unavailable | Reconnect Chrome (connect_browser.sh), then rerun the loop |
| Search creation required | Open Recruiter, create the search using the provided Copilot query, then rerun the loop |
| Confirm search required | User verifies the search is ready, then run with --confirm-search |
| Send confirmation required | Review drafted messages in workbook, get explicit user approval, then run with --confirm-send |
| Page/dialog blocked | Clear the blocker, then rerun the loop |
| Unclear failure | Stop and inspect DEBUG.md; do not manually advance phases |
Review And Send Rules
- Do not rewrite generated drafts unless explicitly asked.
- Do not send InMails without explicit user confirmation.
- Only use
--confirm-sendafter approval. - If approval is missing or ambiguous, stop and ask.
Approved send command:
python3 "$SKILL_DIR/scripts/run_reachout_loop.py" --project "{PROJECT_ID}" --confirm-send
Debugging
Debug commands are intentionally not listed here. For rare troubleshooting after the loop fails, see DEBUG.md.
Debug commands are not a normal workflow. Return to the loop immediately after diagnosis.
Do Not
- NEVER run individual phase scripts to advance a project during normal operation.
- NEVER manually decide the next phase.
- NEVER use ad-hoc
agent-browsercommands to continue workflow progress. - NEVER skip the loop because a phase name looks obvious.
- NEVER edit project state unless explicitly reconciling a known failure.
- ONLY run individual phase scripts when debugging a specific loop failure.
Wrong:
python3 "$SKILL_DIR/scripts/run_extraction.py" --project "{PROJECT_ID}"
python3 "$SKILL_DIR/scripts/run_filter.py" --project "{PROJECT_ID}"
python3 "$SKILL_DIR/scripts/run_enrich.py" --project "{PROJECT_ID}"
Right:
python3 "$SKILL_DIR/scripts/run_reachout_loop.py" --project "{PROJECT_ID}"