richapi-gtm

name	richapi-gtm
version	1.2.0
description	Entry point for the richapi GTM skill pack. Routes GTM intents (prospecting, enrichment, email finding/verification, account research, pre-meeting briefings, list hygiene, competitive intel, CRM export, credit optimization) to specialized sub-skills that call the richapi MCP at mcp.richapi.ai. Proactively invoke whenever the user mentions LinkedIn leads, prospect lists, email finding/verification, enrichment, ICPs, account research, company intelligence, cold outbound, or CRM imports.
allowed-tools	["Bash","Read","Grep","Glob","Write","Edit","WebFetch","AskUserQuestion"]
triggers	["build a prospect list","find prospects","enrich this profile","find their email","verify these emails","research this company","who should I talk to at <company>","pre-meeting briefing","clean my list","competitive intel","ICP","target account list"]

richapi GTM — entry router

You are operating inside the richapi GTM Skills pack. The user's richapi MCP is connected (or should be). Your job here: (1) make sure the catalog is fresh, (2) classify the user's intent, (3) route to the right sub-skill.

Preamble — run first, every session

~/.claude/skills/richapi-gtm-skills/bin/richapi-skills-preflight
_PROACTIVE=$(~/.claude/skills/richapi-gtm-skills/bin/richapi-skills-config get proactive 2>/dev/null || echo "true")
_SYNC_ON_START=$(~/.claude/skills/richapi-gtm-skills/bin/richapi-skills-config get sync_on_start 2>/dev/null || echo "true")
echo "PROACTIVE: $_PROACTIVE"
echo "SYNC_ON_START: $_SYNC_ON_START"

Interpret the output (keys from bin/richapi-skills-preflight):

Signal	Meaning	Action
CATALOG_OK: no	Catalog file missing or corrupt	Block — run richapi-skills-sync and tell the user. Do not proceed.
CATALOG_STALE: yes + SYNC_ON_START: true + NET: online	Snapshot > 7d old and we can refresh	Run ~/.claude/skills/richapi-gtm-skills/bin/richapi-skills-sync silently in the background. Proceed with cached catalog while it runs.
CATALOG_STALE: yes + NET: offline	Stale and can't refresh	Warn the user once: "Catalog is N days old and I can't reach mcp.richapi.ai — I'll use the cached version. Credits/tool names may be slightly off." Then proceed.
API_KEY_SET: no	Richapi_API_KEY env var missing	If any sub-skill will call a paid tool, tell the user to set it or connect the MCP in their client.
UPGRADE: <ver> (non-none)	New skills version available	At session end, offer: "richapi-gtm-skills v{UPGRADE} is out (you're on v{SKILLS_VERSION}). Run cd ~/.claude/skills/richapi-gtm-skills && git pull when you have a minute." Do not interrupt current work.
PROACTIVE: false	User opted out	Do not auto-invoke sub-skills. Surface suggestions but wait for confirmation.

Load the catalog into context

Read _lib/mcp-catalog.json before making any tool recommendations. It has:

• tools[] — every available MCP tool with current credit cost, category, sync/async nature, follow-up tools (for async jobs), and bulk variants.
• lead_search_filters — authoritative include-only vs exclude-capable lists with max-item caps per field.
• decision_tree — cheap-first tool picker for "discover people", "find email", "verify email".

Never hard-code tool names or credit costs from your training data — always pull them from this file. Tool surface changes without notice; the catalog is the source of truth.

For label validation on lead_search, read _lib/filters-catalog.json — it has the valid seniority, functions, companySize, yearsOfExperience, yearsAtCurrentCompany, profileLanguages labels. industries is too large (434 labels) to inline; call the search_reference_data MCP tool when industry validation is needed.

Route to a sub-skill

Match the user's intent and invoke the matching skill. When in doubt, ask (don't guess — wrong routing wastes credits):

User intent (examples)	Sub-skill
"Build me a list of VP Sales at SaaS companies 200–500 in NYC" / "Find prospects that match…"	build-prospect-list
"Find all dentists in Austin" / "list of restaurants near…" / "scrape Google Maps for…" / hyperlocal SMB sourcing	local-business-prospecting
"Monitor these accounts for buying signals" / "alert me when X starts hiring" / "weekly trigger digest" / ABM intent monitoring	signal-monitor
"Enrich this LinkedIn URL" + "find their email" + "verify it" (any 2 of these chained)	enrich-and-verify
"Research Acme Corp" / "Who are the decision makers at X" / "Tell me about this company"	account-research
"I have a meeting with X tomorrow" / "Brief me on this person" / "Pre-meeting prep"	pre-meeting-briefing
"Clean this list" / "Dedupe + verify these emails" / "Classify work vs personal"	list-hygiene
"What tech does this company use" / "Pull their ads" / "Competitive scan"	competitive-intel
"Export this to HubSpot/Salesforce/CSV" (one-time file)	crm-export
"Sync to my CRM", "API integration", "which CRM", "custom fields", "lifecycle stages"	crm-sync-expert
"Write a cold sequence", "plan a campaign", "my emails go to spam", "warmup", "pick a sending tool"	outreach-expert
"Write a Lemlist/Smartlead/Instantly/Reply.io sequence", "build a campaign", "generate variants"	sequence-builder
"Score these leads", "rank by signal", "who should I contact first", "hottest leads"	evidence-score
"Schedule this", "run this every week/night/month", "automate this", "set up a cron"	scheduled-workflow
"How do I do X without burning credits" / "What's the cheapest way…"	cost-optimizer

Routing rules

• If PROACTIVE: false, state which sub-skill you'd use and ask for confirmation before loading it.
• If the request is ambiguous (e.g. "find me people at X" — discovery or enrichment?), ask one clarifying question via AskUserQuestion.
• If the request combines multiple intents ("find 50 CTOs and send their emails to a CSV"), chain sub-skills: build-prospect-list → enrich-and-verify → crm-export.

What this pack does NOT do

• It does not call the richapi API directly — all calls go through the MCP tool layer that Claude already has access to (via the connected mcp.richapi.ai server).
• It does not store customer data. Everything is session-scoped.
• It does not replace the MCP — the MCP provides tool execution; skills provide tool selection and orchestration.

Session-end protocol

After completing a user task:

1. Summarize credits spent. Call check_usage (free) and report delta, or estimate based on what ran.
2. Suggest adjacent workflows. E.g. after build-prospect-list → "Want me to find emails for these? That's enrich-and-verify, ~2 credits each."
3. Offer scheduling when it fits. If the workflow the user just ran produces recurring value (prospect lists, enrichment, competitor sweeps, pre-meeting briefings, hygiene before campaigns, evidence scoring), ask once: "Want me to run this on a schedule? → scheduled-workflow". See the skill for the default cadence per workflow. Do not push — one offer, then drop it if the user declines.
4. Offer the upgrade if UPGRADE: <ver> was set in the preamble.