process-connections

Files (1)

• SKILL.md

SKILL.md (first lines)

## Browser Preflight (MANDATORY — do this FIRST)

Before delegating to any agent, attach the Claude Chrome extension to this session:

1. Tell the user in chat:
   > "I need browser access to LinkedIn. You may see a permission prompt — please approve it. I'll wait."
2. Navigate to `https://www.linkedin.com/feed/` using the Browser tool. This triggers Claude's permission system and attaches the Chrome extension to the session.
3. Confirm the LinkedIn feed page loaded (content visible, not an error/login wall). If it fails, tell the user the extension isn't connected and stop — do NOT delegate to the agent.

If LinkedIn already loaded earlier in this session, this step passes instantly.

Delegate to the **outreach** agent in **process-connections mode**:

This is a multi-session task. The agent works through the user's connections, researching and qualifying ICP matches. It picks up where it left off each session.

## Step 1: Get connections data

**Check if `${CLAUDE_PLUGIN_DATA}/config.json` exists in the plugin data directory.**

**If ${CLAUDE_PLUGIN_DATA}/config.json exists (Pro plan):**
- Fetch unprocessed connections from the dashboard API via a browser tab using `references/api.md`
- If `api_url` is the default `http://127.0.0.1:8001`, open `http://127.0.0.1:8001/health` first so the fetch runs from the HTTP API origin, then run:
  `fetch("API_URL/v1/gtm/connections?status=new&page_size=75&sort_by=created_at&sort_order=asc", {headers: {"X-API-Key": "API_KEY"}}).then(r=>r.json()).then(d=>console.log(JSON.stringify(d)))`
- Users upload their CSV on the dashboard — you don't need to ask for the file

**If ${CLAUDE_PLUGIN_DATA}/config.json does NOT exist (Basic plan):**
- Ask the user for their connections export file (LinkedIn CSV or spreadsheet)
- Read the file — LinkedIn exports include: First Name, Last Name, Profile URL, Email, Company, Position, Connected On

## Step 1b: Check user notes

**For Pro plan connections fetched from the API:** Each connection may have a `notes` field containing user-written instructions or context. Examples:
- "Many prior interactions, say it's been a while and write a personal message"
- "Met at SaaStr conference, reference that"
- "Skip — this is a competitor"

**Always check the `notes` field before processing a connection.** If the note says to skip, mark as skipped. If the note has outreach instructions, follow them when drafting messages. If no notes, proceed with standard ICP qualification.

## Step 2: First-pass exclusion filter (no LinkedIn browsing)

Using only the connection record fields (Position, Company, Industry), immediately exclude clear non-fits before doing any browsing:

- Matches a saved exclusion: competitor companies, excluded industries, "do not contact" personas.
- Role is clearly outside all Tier 1 and Tier 2 ICP roles with no plausible buying influence.
- Company is visibly outside the ICP target with no contrary evidence.

Mark excluded connections `status: "skipped"` and write a one-line `qualification_reason` citing the matched exclusion. Do not research these further.

All remaining connections advance to the research pass.

## Step 3: Deep qualification research pass (max 50-75 per session)

For each connection that passed Step 2, run a two-step research workflow. Track progress in `connection-processing-progress.json`.

### 3a — Visit the person's LinkedIn profile

Confirm:
- Current title and role authority (decision-maker, champion, or IC with no buying influence)
- Current company and tenure
- Location / geography
- Any person-level signals relevant to the ICP (skills, past companies, activity)

Stop researching as soon as a clear exclusion or non-fit is confirmed.

### 3b — Visit the company LinkedIn page

Confirm:
- Industry / category
- Employee count / size signals
- Geography / HQ
- Any product, funding, or growth signals relevant to the pitch

Use the company website **only** if it is clearly linked from the LinkedIn company page. Do not guess domains.

### 3c — Assign ICP tier

Apply saved exclusion rules before positive scoring:

- Saved exclusion industries, companies, personas, or competitors → `status: "skipped"`, include matched exclusion in `qualification_reason`.

Then score against the saved ICP and product/pitch:

- `icp_tier: 1` — clear Tier 1 role (decision-maker / economic buyer) **plus** strong company fit: right industry, company size, geography, and the product pitch maps directly to a problem they own.
- `icp_tier: 2` — adjacent champion, influencer, or partial fit worth keeping: role has buying influence or strong internal access, and company is plausibly in scope even if not a perfect match.
- Not a match — role is IC-only or company is clearly out of scope → `status: "skipped"`.

### 3d — Build qualification fields

**`qualification_reason`** — a compact 2–4 sentence paragraph explaining the ICP decision. Never use a label alone. Must cover:
- ICP fit: role/title vs Tier 1 and Tier 2 criteria, industry, geography, company size, exclusion checks.
- Product/pitch fit: how this person's ownership area maps to the configured outcome, pitch, and messaging.
- Evidence: cite only facts from the connection record, LinkedIn profile, or company LinkedIn page.
- Uncertainty: if evidence is thin, say what is missing — do not invent intent, budget, urgency, or funding.

Examples:
- Tier 1: `Tier 1 fit: VP of Sales at a 200-person B2B SaaS company in the US, matching the saved Tier 1 role and target industry. They likely own pipeline growth and outbound conversion — the exact problem the configured pitch addresses. Company is in the right size band and geography. Outreach should validate current priorities before assuming active budget.`
- Tier 2: `Tier 2 fit: Sales Operations Manager at a mid-market SaaS company. Not a direct economic buyer but holds strong process influence and is often a champion for tooling decisions. Industry and company size align with the ICP; role authority is one level below Tier 1.`
- Skipped: `Not a match: the title indicates an individual contributor with no apparent buying influence for the saved ICP roles. Company industry does not align with the configured product pitch.`
- Excluded: `Excluded: company matches the saved competitor exclusion list. Role fit is irrelevant — this account should not receive outreach.`

**`company_research_summary`** — 2–4 sentences covering: what the company does, industry/category, employee count or size signals, geography/HQ, and any product, funding, or growth signals that are relevant to why this connection is or isn't a fit for the pitch. Source everything from LinkedIn; note if the company page was unavailable.

## Step 4: Store results

Always write results to persistent agent memory first.

**Pro plan (${CLAUDE_PLUGIN_DATA}/config.json exists):**
- Sync results back via browser tab — POST to `API_URL/v1/gtm/sync` with `type: "connections"`
- Every synced record must include: `id` (when present), `linkedin_url`, `icp_tier`, `status`, `qualification_reason`, `company_research_summary`, `notes`, `industry`, `company_size`
- Tier 1/2 matches: set `icp_tier: 1` or `icp_tier: 2`; leave `status` as-is (do not overwrite to skipped)
- Non-matches and exclusions: set `status: "skipped"`, still include `qualification_reason`
- See `references/storage.md` for sync method

**Basic plan (no ${CLAUDE_PLUGIN_DATA}/config.json):**
- Write confirmed ICPs to the user's spreadsheet "ICP Connections" tab
- Columns: Name, LinkedIn URL, Title, Company, Industry, Company Size, Qualification Reason, Company Research Summary, ICP Tier, Status, Date Added
- See `references/storage.md` for spreadsheet method

Deduplicate against existing pipeline data in both cases: for Pro plan, the `status=new` fetch in Step 1 already excludes connections previously processed, so only skip a connection if it also appears in `leads.json` (normalized LinkedIn URL) — sync it as a lead there but do not create a second lead record. For Basic plan, before writing to "ICP Connections", skip any row whose normalized LinkedIn URL already exists in that tab or in the "Leads" tab.

From Kasper-AI/gtm-kit-plugin, MIT.

More from Kasper-AI/gtm-kit-plugin