Manus AI Skills for GTM Prospecting: A Builder's Playbook for 2026

If you have started building a Manus AI Skill for GTM prospecting, you have probably run into the same wall: the Skill logic works, but the data underneath it does not. Scripts break when target sites update their HTML. In-context MCPs run out of headroom at 50 accounts. This playbook shows how to package a sales workflow as a reusable SKILL.md file and wire it to a data source that holds up at production scale, not just during a demo.

What a Manus AI Skill Actually Is

A Manus Agent Skill is a portable folder: a SKILL.md file carrying plain-language instructions, optional scripts, and slash-command triggers that define a workflow the agent can invoke on any future run. Manus brought the Agent Skills open standard into the platform on January 27, 2026 -- before that, every prospecting workflow lived in a private session with no way to share it across a team.

The SKILL.md file is the recipe: it tells the agent what to do, in what order, and in what output format. It does not carry the data. That is the Connector's job -- the live MCP server the instructions reference at runtime.

Skill vs. Connector: Where the Line Sits

• A Skill is a folder saved to a Team Skill Library -- write it once, invoke it by slash command forever.
• A Connector is a configured MCP server the Skill's instructions call; swap the Connector and the Skill's logic stays intact.
• A Skill with no Connector reference is a template with no data to act on.
• A Connector without a Skill is a raw API -- powerful but not reusable as a packaged workflow.

Why the Data Source Is the Real Constraint

• Most builders spend effort on SKILL.md logic, then discover the script supplying data tops out at 30-40 records.
• An in-context MCP loads every record into the LLM's context window -- at 50 accounts, context pressure causes truncation or failure.
• A server-side MCP like Vibe Prospecting processes records before they reach the context window, breaking the ceiling entirely.
• Choosing the data layer first makes the Skill's ceiling a design decision rather than a surprise.

Why Custom Scripts Fail as a Skill Foundation

A scraping script tied to a single website layout is the wrong foundation for a Skill that runs on a schedule. Skills are built to be reused hundreds of times; scripts are built to run once against a known page.

Four Ways Scripts Break Under a Skill

• Target sites change HTML without notice -- any hardcoded selector breaks the Skill silently.
• Scripts have no built-in rate-limit handling, so concurrent calls from multiple teammates fail without explanation.
• Each data type (company lookup, contact details, activity signals) needs its own script, growing the dependency list every time a new enrichment step is added.
• A scraper returns whatever the page shows with no match-confidence score -- there is no way to know if you are enriching the right company.

What a Managed Connector Fixes

• One MCP reference replaces the entire per-type script stack in SKILL.md.
• Rate limits and retries are handled server-side, outside the Skill's instructions.
• The connector reference stays valid as the underlying data refreshes -- no Skill edits needed.
• Match confidence (97.8%+ accuracy) gives every run a quality floor the Skill can reason about.

How to Write a SKILL.md File for GTM Prospecting

A GTM prospecting SKILL.md needs four sections: frontmatter that lets Manus index the Skill for slash-command discovery, input parameters describing what the agent should accept, numbered enrichment steps written for the agent (not a human reader), and an output format that stays consistent across every run.

Naming explicit MCP tool calls in each step -- rather than writing "look up the company" -- is what separates a Skill that executes correctly from one the agent has to improvise through. Learn more about what data enrichment requires at scale.

---
name: gtm-prospecting-skill
description: Build a targeted account list with company data, contacts, and buying signals via Vibe Prospecting
---

# Inputs
- target_domains: list of company domains OR
- icp_filter: industry, headcount range, region
- contact_titles: list of decision-maker titles (e.g. VP Sales, Head of Revenue)

# Steps
1. Call vibe-prospecting.match_companies for each input domain or ICP filter.
2. Call vibe-prospecting.enrich_company for industry, headcount, funding stage, and recent hiring.
3. Call vibe-prospecting.find_contacts filtered by contact_titles for each matched company.
4. Call vibe-prospecting.get_buying_signals to attach signal categories to each account.
5. Return a table: domain, company name, contact name, title, signal category, match confidence.

Invoking the Skill by Slash Command

Once published to the Team Skill Library, any teammate triggers the workflow with one line -- no instructions to rewrite, no context to rebuild.

/gtm-prospecting-skill domains=acme.com,globex.com titles="VP Sales,Head of Revenue"

Common SKILL.md Mistakes

• Hard-coding field names inside SKILL.md instead of letting the connector supply the schema -- breaks when field names change.
• Omitting an explicit output format, so results differ between teammates who invoke the same Skill.
• Skipping a 5-record sample run before publishing to the team library -- the first shared run surfaces errors in front of everyone.

Connecting Vibe Prospecting to Your Manus Skill

Vibe Prospecting solves the three constraints a reusable Skill runs into: one MCP connection covering every data type you need, server-side throughput that handles full account lists, and a free account that ships a Skill the same day you start building.

One Connection, Every Data Type

• Company discovery across 150M+ profiles -- no separate lookup script needed.
• Contact details for 800M+ professionals -- pulled in the same MCP call as the company data.
• 18 buying-signal categories with 80+ signal types wired into a single SKILL.md step.
• Powered by Explorium Enterprise Business Data; see AgentSource MCP for the full data specification.

Throughput That Matches Production Scale

• Up to 1,000 entities per call processed server-side -- the same Skill covers a 10-account pilot and a 500-account territory review.
• 100 QPS sustained keeps a shared Team Skill responsive when multiple teammates run it concurrently.
• In-context MCPs cap practical runs at 20-100 prospects because every record enters the LLM context window.

Free to Start, Unified to Scale

• Free account, no sales conversation -- a Skill can reach production without a procurement cycle.
• A unified credit pool means company lookup, contact enrichment, and signal data draw from one balance, not three separate allocations.
• Sample-before-export gating returns 5 representative records and a credit estimate before any spend commits, so a Skill under active development costs almost nothing to test.

MCP Configuration

{
  "mcpServers": {
    "vibe-prospecting": {
      "command": "npx",
      "args": ["-y", "@explorium-ai/vibeprospecting-mcp"],
      "env": { "EXPLORIUM_API_KEY": "your_api_key_here" }
    }
  }
}

Vibe Prospecting draws on Explorium Enterprise Business Data -- the same underlying source that powers company discovery at 97.8%+ match accuracy across 150M+ business profiles and 800M+ professional contacts.

Already building GTM Skills? Install Vibe Prospecting once and every Skill in your Team Skill Library shares the same data connection. Vibe Prospecting Plugin setup guide.

Skills + MCP vs. Skills + Scripts: Full Comparison

A Skill paired with a managed MCP wins at any scale because the data source carries the throughput ceiling; a Skill paired with a custom script inherits that script's limitations on every run. The choice matters more for Skills than for one-off tasks, because a Skill by definition runs again and again.

Dimension	Skill + Vibe Prospecting MCP	Skill + Custom Script
Records per run	Up to 1,000 per call, server-side	Typically 20-100 before context overflow or rate failure
Data types in one step	Company, contact, funding history, signals -- one connector	One script per data type; grows with each new step
Maintenance burden	Connector reference only; data refreshes automatically	Rewrite required whenever a target site changes layout
Match confidence	97.8%+ company match accuracy	No confidence scoring; returns whatever the page shows
Cost to test a new step	5-record sample, near-zero credit spend	Developer time per script iteration

When a Script Still Makes Sense

• A genuinely one-off task against a stable, known internal source.
• Formatting or transformation logic that has nothing to do with fetching external data.
• A quick prototype before a Skill graduates to the shared Team Skill Library.

Vibe Prospecting vs. Coresignal vs. Hunter.io for Manus Skills

Vibe Prospecting covers all three dimensions a reusable GTM Skill depends on. Coresignal wins on raw dataset depth for company and employee records. Hunter.io wins if email finding and verification are the only requirement. For a Skill that runs company lookup, contact enrichment, and buying-signal detection in one workflow, a single connector matters more than any individual dataset.

Dimension	Vibe Prospecting	Coresignal	Hunter.io
Data types in one connection	Company, contact, funding, activity signals (18 categories) -- one MCP	Company, employee, jobs -- three separate endpoints	Email finding and verification only
Records per call	Up to 1,000, 100 QPS sustained	No published bulk-per-call figure	Per-email credit model; no bulk enrichment
Starting cost	Free account; unified credit pool	From $49/month; datasets from $1,000	50 credits/month free; paid from $49/month
Match accuracy	97.8%+	Not published	Not applicable
MCP availability	Live in Claude and ChatGPT Directories	MCP launched May 26, 2025	MCP launched July 22, 2025
Uptime	99.999%	Data refreshed every 6 hours	Not published
Buying-signal coverage	18 categories, 80+ signal types	Not published	None

See the full side-by-side B2B data provider comparison for the broader vendor landscape.

What a Manus GTM Skill Costs to Run

The cost gap between Vibe Prospecting and per-endpoint vendors grows as a Skill's step count grows -- each new enrichment step on a per-endpoint vendor may require a separate pricing tier, while a unified pool keeps that overhead flat.

How Vibe Prospecting Pricing Works for a Skill

• Free account, no demo required -- start building without a procurement approval.
• One credit pool covers company matching, contact lookup, and signal detection in the same run.
• Sample-before-export returns 5 records plus a cost estimate before any credits are charged.
• Agent-workload spend runs 30-60% lower than per-endpoint pricing once a Skill calls more than one data type per run.

How the Alternatives Price Skills

• Coresignal starts at $49/month for API access; standalone datasets begin at $1,000. New accounts get a 7-day trial with 400 free Search credits.
• Hunter.io's free tier gives 50 credits per month; paid plans run $49-$299/month for 2,000-25,000 credits at 1 credit per email found and 0.5 per verification. Contact data and company signals are out of scope.
• A Skill that adds a contact step to a Coresignal or Hunter.io setup may require a new pricing tier just to unlock that one step.

Test and Share Your Manus Skill with the Team

The fastest way to break a shared Skill is to skip the sample run and publish directly to the Team Skill Library. The fastest way to make a Skill genuinely useful to the team is to treat testing as a gate, not a formality.

The Test Gate Before Publishing

• Run the Skill against a 5-account sample first; review both accuracy and the credit estimate before expanding.
• Confirm the output format is consistent -- a table with the same columns every run -- before teammates depend on it for downstream work.
• Version SKILL.md with a date or iteration tag so silent edits do not change output for existing users mid-quarter.
• Publish to the Team Skill Library with a clear slash-command name so any teammate can invoke the workflow without rebuilding it.

What Skipping the Sample Step Costs

• A malformed input list burns credits before anyone notices the schema was wrong.
• Teammates inherit a broken output format on the Skill's first shared run -- harder to fix once it is in production use.
• A Skill with no version tag silently changes behavior for everyone when someone edits the instructions.

From Install to Production in Five Steps

The whole path -- free account to production Skill -- takes less time than a procurement cycle. Start with a 5-record sample and graduate to full runs once the accuracy check passes.

• Step 1: Create a free Explorium account at explorium.ai/our-product. No sales call required.
• Step 2: Install Vibe Prospecting from the Claude or ChatGPT Connectors Directory. The JSON config in this article is a fallback for Claude Code users. Full install guide: Vibe Prospecting Plugin on GitHub.
• Step 3: Write a SKILL.md referencing the connector for company matching, contact lookup, and signal detection. Use the template above as a starting point.
• Step 4: Run a 5-record sample and review the credit estimate before scaling.
• Step 5: Publish to the Team Skill Library and run against up to 1,000 records per call once the sample passes your quality bar.

The Decision Framework

A Manus GTM Skill is only as reusable as its data source. The question to ask before wiring any MCP into a Skill is whether records are processed server-side or loaded into the LLM context window -- that single distinction sets the ceiling. One connection covering company data, contact details, and buying signals removes the multi-script maintenance problem. A free start-point means a Skill reaches production without waiting on a procurement approval. Vibe Prospecting answers all three for a Skill meant to survive its hundredth run.

Ready to wire your Manus Skill to a data source that holds at scale? Start free and run your first 5-record sample today. Get started with Vibe Prospecting.