Outreach Tracker

Drive the outreach follow-up lifecycle from Mercury's relationship memory: find what needs action today, draft each message/withdrawal/close, and apply the state change once the user approves. This is the companion to recruiter-outreach (which initiates outreach and records attempts); the tracker maintains them.

Core principle: draft + ask, never auto-send

require_send_consent is on by default. This skill drafts messages and proposes actions; the human approves each one before anything is sent or withdrawn on LinkedIn. Do not send, message, or withdraw without explicit per-action consent.

Prerequisites

• Mercury CLI — mercury outreach due|update|blocked|budget
• LinkedIn MCP — get_inbox, get_conversation (reply detection), send_message (approved nudges)
• Chrome MCP — for invitation withdrawal (no MCP withdraw tool exists)

Workflow

1. Find what's due

mercury outreach due

Each line is one of three action kinds with the person, company, and reason:

• [WITHDRAW] — an invite has gone unaccepted past the threshold (~7d).
• [FOLLOWUP] — they accepted but haven't replied to your first message (~4d).
• [CLOSE] — your follow-up went unanswered past the grace window (~7d).

Also glance at the InMail budget so you can factor credit cost into any InMail-based follow-ups:

mercury outreach budget

2. Detect replies FIRST (before acting on anything)

A reply rescues an attempt from any cadence — never nudge or close someone who already responded. For each due item (and any accepted/followed_up attempt), check the conversation:

get_conversation(linkedin_username = "{username}")   # or search_conversations / get_inbox

If they replied after your last message, mark it engaged and stop automation for that person:

mercury outreach update --id {attemptId} --state engaged --reason "replied: {gist}"

Marking engaged also refunds an InMail credit if one was spent within the 90-day window — the budget updates automatically.

3. Handle each remaining due action (with consent)

[FOLLOWUP] — accepted, no reply: Draft a short, friendly nudge (≤ ~400 chars, specific, no guilt-trip). Show it to the user. On approval, send and record:

send_message(linkedin_username = "{username}", message = "{approved text}", confirm_send = true)
mercury outreach update --id {attemptId} --state followed_up --reason "gentle nudge sent"

[WITHDRAW] — invite ignored: Explain you'll withdraw the pending invitation (recalling the note). The LinkedIn MCP has no withdraw tool, so drive the browser with Chrome MCP:

chrome: navigate_page("https://www.linkedin.com/mynetwork/invitation-manager/sent/")
chrome: take_snapshot            # locate the row for {Name}
chrome: click(<the person's "Withdraw" button>)
chrome: handle_dialog(accept)    # confirm if prompted

Then record the outcome (this also company-blocks them so they're skipped in future recruiter-outreach runs):

mercury outreach update --id {attemptId} --state invite_ignored --reason "invite withdrawn after {N}d"

Or use mercury outreach withdraw --id {attemptId} — it marks the block and, if the browser step couldn't be confirmed, degrades gracefully (the invite may still linger on LinkedIn; the block bookkeeping still happens).

Caveats to tell the user: LinkedIn blocks re-inviting a withdrawn person for ~3 weeks; weekly invite limits still apply; sent notes can't be edited (only withdrawn).

[CLOSE] — follow-up exhausted: No send needed. Mark unresponsive (company-blocks them):

mercury outreach update --id {attemptId} --state unresponsive --reason "no reply after follow-up"

4. Summarize

Present a short table of what was done and what's pending approval:

Note the updated InMail budget (credits remaining vs reserve floor) if any InMail action was taken or refunded.

Notes

• A block is scoped to (person, company URN) and expires after the configured cooldown (default 9 months) — the tracker doesn't permanently burn a contact.
• Reply detection is best-effort; if the inbox tools can't confirm a reply, fall back to asking the user "did {Name} reply?" before nudging or closing.
• This skill is safe to run on a daily schedule: with consent gating on, the worst case is a list of drafts awaiting your approval.