/outreach

OUTREACH (cold sweep at a company)

The outreach flow does two things at a target company:

1. **Send connection requests** to 2nd-degree contacts (recruiters, peers, leads). Always "Send without a note" — LinkedIn rate-limits personalized invites. 2. **Send a first DM** to your existing 1st-degree connections at the company.

Out of scope (you handle these manually):

• ❌ Replying to DMs once a thread is warm
• ❌ Follow-up nudges on stale outreach
• ❌ Sending a resume / file attachments

---

Step 1 — Pick a company

**If the command was called with an argument** (`/jobs outreach Mixpanel`), use that company. Skip the rest of this step.

**Otherwise**, read `job_tracker.csv` for outreach candidates: rows where `Referral Needed=YES` AND `Referral Status` ∈ `Outreach Pending` / `Connection Pending`.

• **Has candidates** → auto-pick the most urgent one: smallest `Days Left` (Referral Deadline − today; negative = overdue first). Tiebreaker: oldest `Discovered Date`. Print:

> `Picked: <Company> — <Role> (deadline in <N> days)`

Proceed to Step 2 immediately. No confirmation prompt. (If you want to override, stop the run and re-invoke as `/jobs outreach <OtherCompany>`.)

• **No candidates in the tracker** → ASK the user: "Tracker has no companies queued for outreach. Which company do you want to run outreach on? (You can also run `/jobs find` first to discover jobs.)" Wait for a company name, then proceed with that.

If the picked company is `Connection Pending`, jump to **Step 4 — Check pending acceptances** instead of Step 2.

---

Step 2 — Find contacts (1st-degree FIRST, then 2nd)

LinkedIn via Chrome:

1. Navigate to `https://www.linkedin.com/company/<slug>/people/`. Extract the numeric company ID by scanning the page for `urn:li:fsd_company:<id>` and picking the most-frequent match.

2. **Step 2A (MANDATORY): 1st-degree only**, no keyword filter:

   https://www.linkedin.com/search/results/people/?currentCompany=["<id>"]&network=["F"]

Read every result. These are your existing connections at the company and are always the top picks for direct DMs.

3. **Step 2B: 2nd-degree fallback**, with role-relevant keywords:

   https://www.linkedin.com/search/results/people/?currentCompany=["<id>"]&network=["S"]&keywords=recruiter OR <role keyword>

Use role keywords inferred from the user's resume(s) in `resumes/` (e.g. "data scientist" / "engineer" / "product manager").

4. **Never combine `network=["F","S"]` in one search.** LinkedIn sorts combined results by keyword relevance, which buries 1st-degree contacts below 2nd-degree ones. Always do the two searches separately.

5. URL params are `network` and `currentCompany`, NOT `facetNetwork` / `facetCurrentCompany`.

Extract per contact: name, title, connection degree, mutual-connection count, profile URL.

**Priority order:**

1. **Any 1st-degree connection** at the company — always beats 2nd-degree, regardless of title. Goes through Step 3 (DM flow). 2. **2nd-degree recruiter / Talent Acquisition.** Connection request via Step 4. 3. **2nd-degree peer** (data scientist, engineer, product, etc. — match on the user's role from their resume in `resumes/`). Connection request via Step 4. 4. **2nd-degree lead / manager** in the user's function. Connection request via Step 4.

Show two ranked tables, one per network tier:

1st-degree (DM directly):
# | Name | Title | LinkedIn URL | Mutuals

2nd-degree (connection request):
# | Name | Title | LinkedIn URL | Mutuals

---

Step 2C — Connection-request quota (per company)

Use a per-company quota based on how many 1st-degree connections the user already has there:

quota = max(0, (10 − count_1st_degree) × 5)

| 1st-degree available | Connection requests to send | |---|---| | 10+ | 0 (already have plenty of internal contacts) | | 9 | 5 | | 8 | 10 | | 7 | 15 | | 5 | 25 | | 2 | 40 | | 1 | 45 | | 0 | 50 |

**No global weekly cap is enforced.** Send the per-company quota in full. The only stop signal is LinkedIn itself — see Step 4B for the CAPTCHA / rate-limit handling.

---

Step 3 — DM existing 1st-degree connections

For each 1st-degree contact selected:

A. Open profile and verify connection level

Navigate to the contact's profile URL. Confirm they show as 1st connection.

B. Prior-thread check (MANDATORY before drafting)

Click Message to open the chat overlay. Scan the thread history:

| Thread state | Action | |---|---| | No prior messages | Proceed to **C (draft message)** | | Prior message from user, contact replied | Skip — warm thread, user handles manually | | Prior message from user, no reply, ≥7 days ago | **SKIP** — do not re-engage. Log `Skipped (prior unanswered DM <date>)` | | Prior message from user, no reply, <7 days ago | Skip for now (still in reply window). Log and move on |

Close the dialog if skipping.

C. Draft and send the first DM

**Step 1: Build the pitch from the user's resume.** Read whatever is in `resumes/`. Extract:

• The user's **first name** (top of the resume)
• The user's **current title** + **employer** (most recent role)
• **Years of experience** (from the experience timeline)
• **Top 2-3 specializations** (from the headline / summary / repeated keywords)

Construct a single first-person sentence about the user. Examples of the right shape: > "I'm a senior data scientist with 5+ years in experimentation and product analytics, currently at <Employer>." > "I'm a staff ML engineer with 8+ years building production LLM systems, currently at <Employer>."

Don't invent details not in the resume.

**Message style rules (strictly enforced):**

• Short: 2-3 sentences max
• Conversational, not corporate
• No em-dashes (—) anywhere
• No filler phrases like "I hope you're doing well"
• No "Best, <name>" sign-off — end naturally
• **First person throughout** — the DM is sent from the user's account, so it must read as the user speaking. Never use third person ("He is...", "She is...") in any outgoing mes