keinsaasforever/gtm-pipeline-skills/gtm-setup

GTM Pipeline Setup

Welcome. This skill walks you through everything needed to start using the GTM pipeline skills — from installation to your first client ICP.

Read `~/.claude/skills/gtm-pipeline/_shared/conventions.md` before executing.

When to Use

• First time after cloning the repo
• Adding a new API key or PhantomBuster agent
• Setting up a new client (use Step 4 only)
• Troubleshooting a missing config value

Step 1 — Verify Installation

Check that the skills are symlinked into Claude's skills directory.

ls ~/.claude/skills/ | grep gtm

Expected output: 10 directories — gtm-setup, gtm-company-search, gtm-company-enrichment, gtm-contact-filter, gtm-demo, gtm-outreach, gtm-people-enrichment, gtm-people-search, gtm-pipeline, gtm-signal-search.

If any are missing:

• Navigate to the repo root
• Run ./install.sh
• Re-run the check above

Confirm with the user before continuing.

Step 2 — Environment Variables

Check whether an env file exists and which keys are already set.

Where to look

1. Check ~/.claude/skills/gtm-pipeline/_shared/local.md for GTM_ENV_PATH
2. Default location if not set: ~/.env.gtm
3. If neither exists: offer to create ~/.env.gtm now

Required keys

Walk through each key. For each one, check if it's already set in the env file. Only prompt for missing ones.

Optional: If user won't use PhantomBuster (no LinkedIn automation), skip PHANTOMBUSTER_API_KEY, LINKEDIN_SESSION_COOKIE, LINKEDIN_USER_AGENT. OPENROUTER_API_KEY is not needed by default — signal-search's default agent backend has the Claude agent score in-context (see signal-search → Model Routing).

For `LINKEDIN_SESSION_COOKIE`: Tell the user to open LinkedIn in Chrome, go to DevTools → Application → Cookies → li_at value.

For `LINKEDIN_USER_AGENT`: the exact user agent string of that same browser (DevTools console: navigator.userAgent). PB scripts require it alongside the cookie.

Minimum viable set (to run a basic demo): PIPE0_API_KEY, FULLENRICH_API_KEY, SERPAPI_API_KEY, PARALLEL_API_KEY, FIRECRAWL_API_KEY. (No LLM key needed — the agent scores signals in-context. OPENROUTER_API_KEY only if you opt into the legacy openrouter signal backend.)

Writing keys

If the env file doesn't exist yet:

touch ~/.env.gtm

For each missing key, append it:

echo 'KEY_NAME=value' >> ~/.env.gtm

Then set GTM_ENV_PATH in _shared/local.md:

GTM_ENV_PATH=/Users/{username}/.env.gtm

Step 3 — PhantomBuster Agent IDs

Only needed if user will use PhantomBuster for LinkedIn automation.

Option A — Auto-lookup via PhantomBuster MCP

If the PhantomBuster MCP is available, use PHANTOMBUSTER_GET_AGENTS_FETCH_ALL to list all agents. Match agent names to config keys:

Write all matched IDs to _shared/local.md:

## PhantomBuster Agent IDs
 
| Config Key | Agent ID |
|------------|----------|
| PB_AGENT_CONNECT | 123456789 |
| PB_AGENT_SN_ACCOUNT | 987654321 |
...

Option B — Manual

If PB MCP is not available, show the user _shared/local.example.md and instruct them to:

1. Log in to PhantomBuster dashboard
2. Find each agent and copy its ID from the URL or settings panel
3. Fill in _shared/local.md (copy from local.example.md)

Skip entirely

If user doesn't plan to use PhantomBuster, skip this step.

Step 4 — First Client Setup

Ask: "Do you want to set up your first client now? If yes, provide the client's domain (e.g. acme.com)."

If user declines, skip to Step 5.

4.1 Create client directory

Ask for a client slug (e.g. acme for acme.com). Create the working directory:

{slug}-gtm/
├── context/
├── prompts/
└── csv/
    ├── input/
    ├── intermediate/
    └── output/

4.2 Auto-research the domain

Before asking any questions, research the client's domain automatically. Run three sources in parallel:

Firecrawl — scrape homepage + about/careers pages: Use firecrawl_scrape or firecrawl_crawl on the domain. Extract:

• What they sell / core product or service
• Target customer (explicit or implied)
• Industries served
• Key value propositions
• Company size / funding stage signals
• Tone of voice from copy

SerpAPI — recent news and announcements: Search: "{company name}" OR site:{domain} with recent date filter. Extract:

• Funding rounds
• Company description from third-party sources
• Customer segments mentioned in press

Parallel web search (if Parallel API key set): Query: "{company}" {domain} — what they sell, who they target, key customers, industries served

4.3 Draft profile.md and icp.md immediately

Do not wait for user input. Draft both files from research findings now. Mark inferred fields with [INFERRED].

Draft `context/profile.md`:

# {Company Name} — Client Profile
 
## What They Sell
{product/service description from research}
 
## Target Buyer
{who buys from them — role, company type, industry}
 
## Value Proposition
{how they describe their own value}
 
## Reference Customers / Segments
{any named customers or segments found}
 
## Tone of Voice
{formal/casual, technical/plain — inferred from website copy}
 
## Notes
-Profile auto-researched from {domain} on {date}
-Fields marked [INFERRED] should be confirmed with the client

Draft `context/icp.md`:

# {Company Name} — ICP Definition
 
## Company ICP
 
### Target Industry
{industries from research} [INFERRED]
 
### Company Size
-Min employees: [INFERRED — ask to confirm]
-Max employees: [INFERRED — ask to confirm]
 
### Location
{geography signals from research} [INFERRED]
 
### Revenue Range
[INFERRED — ask to confirm]
 
## Contact ICP
 
### Target Job Titles / Roles
{roles from research} [INFERRED]
 
### Seniority
[INFERRED]
 
## Contact Filter Configuration
 
### Job Tier Definitions
[To be filled after Step 5 refinement questions]
 
### Industry Tiers
[To be filled after Step 5 refinement questions]
 
### Location Tiers
[To be filled after Step 5 refinement questions]
 
### Company Size Filter
[To be filled after Step 5 refinement questions]
 
### Sort Order
job_tier → industry_tier → comp_loc_tier → person_loc_tier → employee_count_missing → icp_kw_score (desc)

Show the user what was auto-filled vs. inferred, then ask 5 refinement questions:

4.4 Five Refinement Questions

1. Who are your hardest-to-replace customers?

(This defines your real ICP — not who you want to sell to, but who gets the most value and keeps paying)

1. Who should we NOT target? (Anti-ICP)

(Industries, company types, roles, or geographies that are a bad fit, take too long to close, or drain support)

1. Which job titles actually convert?

(Not who answers the phone — who signs the contract or champions the deal internally)

1. What's your geographic priority?

(Which countries/regions come first? Are there any we should skip entirely?)

1. What tone do you use in outreach?

(Formal or casual? Any message examples you love or hate? Languages?)

4.5 Update files with refinement answers

After collecting answers, update context/profile.md and context/icp.md with confirmed values. Remove [INFERRED] markers where confirmed. Fill in the Contact Filter Configuration section in icp.md.

Step 5 — Skill Tour

Show available skills with a one-line summary:

## GTM Pipeline Skills
 
/gtm-pipeline:setup            You are here — onboarding and first client config
 
/gtm-pipeline:pipeline         Orchestrator — plan and run the full pipeline end-to-end
 
/gtm-pipeline:company-search   Build a company list from Sales Navigator, Parallel FindAll,
                                Firecrawl Agent, or web scraping
 
/gtm-pipeline:company-enrichment  Enrich companies with headcount, revenue, growth metrics.
                                   Run ICP scoring (0–100) against your icp.md
 
/gtm-pipeline:signal-search    Find buying intent signals (funding, hiring, leadership changes).
                                Scores each signal 1–100 for purchase intent
 
/gtm-pipeline:people-search    Find contacts at target companies by job title, or discover
                                people by persona (no company list needed)
 
/gtm-pipeline:contact-filter   Classify and rank contacts before enrichment — saves credits
                                by rejecting non-ICP contacts early
 
/gtm-pipeline:people-enrichment  Enrich contacts with verified work email and phone number
 
/gtm-pipeline:outreach         Run LinkedIn connection requests and personalized messages
                                via PhantomBuster — one profile at a time
 
/gtm-pipeline:demo             Generate a demo lead list (~10 enriched contacts + messages)
                                from an ICP description — no company list needed

Suggested first run:

/gtm-pipeline:demo

Runs a mini end-to-end pipeline for 10 contacts. Good way to verify keys are working and see the output format before committing credits to a full run.

For a full pipeline:

/gtm-pipeline:pipeline

Assesses your starting data, picks Company-First or Signal-First workflow, estimates costs, and walks you through execution.

Configuration Reference

Files written by this skill

Re-running setup

This skill is safe to re-run at any time:

• It only appends missing keys — won't overwrite existing ones
• It will update local.md agent IDs if PB MCP is available
• It will ask before overwriting existing profile.md or icp.md

What's Missing (To Document)