firsttouch-agent-skill-packs

FirstTouch Revenue Skills - Customer Pack

Turn Claude, Cursor, Codex, and ChatGPT into FirstTouch-native LinkedIn revenue operators.

This pack gives customers a structured set of plays - reusable skills that execute real LinkedIn-outreach workflows with FirstTouch, human approvals, account-safety limits, and HubSpot context when available.

It is built to the open Agent Skills standard (agentskills.io), so every skill works in Claude Code, Claude.ai, Cursor, Windsurf, and any MCP-compatible harness.

Why this pack exists

Most "AI outreach" tools stop at generating copy. Customers can get an agent to write a message, but not to:

• act on a warm signal like post engagement, a form fill, a deal stall, or a website visit
• find the right person on LinkedIn and reach out safely
• prepare approval-ready FirstTouch queues instead of autonomous sends
• log and route through HubSpot when CRM access is connected
• respect account-safety limits and human approval gates

Each play closes that loop. This is the difference between an agent that sounds like an SDR and one that operates like one.

How the pack is organized

firsttouch-skills/
├── README.md                              ← you are here
├── references/
│   ├── onboarding.md                      ← first-run questions, play catalog, account limits
│   ├── system-grounding.md                ← how FirstTouch + HubSpot + agents fit together
│   ├── mcp-setup.md                       ← connect HubSpot MCP, FirstTouch MCP, enrichment MCPs
│   ├── messaging-framework.md             ← FirstTouch LinkedIn messaging methodology
│   ├── safety-governance.md               ← approval gates, send safety, account limits
│   └── troubleshooting.md                 ← MCP failures, cooldowns, recovery
└── skills/
    ├── firsttouch-messaging/              ← foundation: how to write outreach
    ├── warm-engager-followup/             ← social engagement flow
    ├── social-campaigns/                  ← one-time static-template campaigns
    ├── icp-outbound-builder/              ← AI SDR
    ├── founder-led-outbound/              ← founder AI SDR
    ├── inbound-speed-to-lead/
    ├── hubspot-signal-to-linkedin-touch/
    ├── website-visitor-followup/
    ├── stalled-deal-reactivation/
    ├── customer-referral/
    ├── hubspot-social-task-runner/        ← execute due HubSpot social tasks
    ├── team-performance-report/           ← FirstTouch team metrics + attribution
    ├── sequence-qa-reviewer/
    ├── workspace-audit/
    ├── campaign-pause-and-fix/            ← mid-send escape hatch: pause, fix, restart
    ├── founder-post-generator/            ← posts that create the engagement to harvest
    ├── post-demo-followup/                ← same-day demo follow-up + stakeholder expansion
    └── champion-job-change/               ← champion moved: new-company path + old-account protection

Two layers:

1. Foundation (references/ + firsttouch-messaging) - shared grounding every play assumes.
2. Customer plays - self-contained, revenue-tied workflows. Governance and attribution rules live in shared references instead of separate customer-facing plays.

Get started in 5 steps

Claude Code users: skip the steps below - install your pack as a plugin and the FirstTouch MCP connects automatically:

/plugin marketplace add First-Touch-Inc/firsttouch-agent-skill-packs
/plugin install founder-pack@firsttouch        # or ae-pack / bdr-pack / revops-pack

Then run /mcp once to authorize the FirstTouch server (a "Needs authentication" status before that is expected), answer the onboarding questions when the agent asks, and run your first play. Everyone else:

1. Download your pack (below) - or grab individual skills from packs/skills/.
2. Install it for your agent - platform guides in install/: Claude Code · Claude.ai · Cursor/Windsurf · ChatGPT.
3. Connect the FirstTouch MCP (https://mcp.firsttouch.ai) - required by every play. HubSpot MCP is optional but unlocks the CRM plays. See references/mcp-setup.md.
4. Answer the 3 onboarding questions in references/onboarding.md - LinkedIn account type, HubSpot access, which play to run first.
5. Run your first play and approve the drafts - nothing sends without your approval.

Download your pack

Single skills for Claude.ai (one skill per zip, references bundled): packs/skills/

Install (for end customers)

1. Drop any skill folder into your agent's skills directory:
   • Claude Code: ~/.claude/skills/<skill-name>/ (or .claude/skills/ per-project)
   • Claude.ai: Settings → Features → upload a single-skill zip from packs/skills/
   • Cursor / Windsurf / custom: copy the folder anywhere the agent can read
2. Connect the MCPs your plays need - see references/mcp-setup.md.
   • FirstTouch MCP is required for FirstTouch execution and approvals.
   • HubSpot MCP or a service key/private app token unlocks HubSpot-specific plays.
   • If HubSpot is unavailable, run FirstTouch-only plays such as social engagement, social campaigns from Discover Contacts or imported lists, AI SDR from Discover Contacts, founder AI SDR, sequence QA, and workspace audit.
3. Complete references/onboarding.md before running the first play.
4. Start with high-intent plays first; add outbound only after those are running and account limits are respected.

First-run onboarding and recommended rollout

Every pack should ask three questions before running volume:

Recommended order:

1. For founders, start with Social Engagement Flow first. It does not require HubSpot and uses the warmest signal available.
2. Run other high-intent plays next: inbound speed-to-lead, website visitor follow-up, HubSpot signal touches, customer referral thank-you, and stalled-deal reactivation when the required data exists.
3. Add one-time social campaigns when the audience is narrow and the user can approve static flow templates.
4. Add AI SDR / Founder AI SDR only after warm motions are running, with daily approval batches of 10 free/basic or 20 Sales Navigator/Premium contacts.

The plays that drive the most value

1. Social Engagement Flow - act on post likes/comments first; for founders this is the number one starter play and can use owned posts or a relevant competitor founder/influencer profile.
2. Social Campaigns - build a narrow campaign using either row-level dynamic approvals for rep/BDR pushes or static templates with flow-level approval for governed one-time campaigns.
3. AI SDR / ICP Outbound - build daily approval-ready batches from a HubSpot list or FirstTouch Discover Contacts.
4. Founder AI SDR - the AI SDR workflow in founder voice, with the same approval queue and recommended 10/20 daily connection caps.
5. Inbound Speed-to-Lead - attach LinkedIn connection and follow-up to signups, trials, or demo requests.
6. Website Visitor Follow-Up - act on pricing/demo/product-page intent from HubSpot tracking or RB2B/list sources.
7. HubSpot Signal Touches - turn lifecycle, list, or deal events into timely social touches.
8. Stalled Deal Reactivation - trigger a contact-based HubSpot workflow for contacts associated to open deals not Closed Won/Lost with no engagement for 60+ days, then draft a fresh approved LinkedIn touch.
9. Customer Referral Thank-You - connect with new customers, thank them, ask for feedback, and invite light referrals.
10. Sequence QA / Workspace Audit - make sure the setup and messaging are safe before scaling.

Plays at a glance

Design principles (Anthropic Agent Skills standard)

• Progressive disclosure - only the relevant context loads. Metadata is always present; the body loads on trigger; references load on demand.
• Passive, deliverable-oriented - each skill defines a deliverable, not a goal. The agent produces an artifact; a human approves before anything sends.
• Human-in-the-loop by default - the agent never sends or publishes a flow without chat approval. Dynamic plays require per-send approval; one-time social campaigns can use flow-level approval for the exact audience and static templates. FirstTouch also offers an optional in-product approval layer (off by default) that pauses sends as approval tasks in FirstTouch or HubSpot - recommended for any flow that keeps running after the chat ends (see references/safety-governance.md).
• HubSpot-aware, not HubSpot-blocked where possible - HubSpot-specific plays say so clearly; AI SDR and founder AI SDR can run from FirstTouch Discover Contacts when no HubSpot list exists.

Persona Packs

In addition to the flat skills/ folder above, this repo ships 4 downloadable persona packs - self-contained bundles that collect the right skills for a specific role.

Shared-core model

All 4 packs include the same 6 core plays, each with a persona-specific lens:

1. Inbound speed-to-lead - fast LinkedIn touch on signups, trials, demo requests
2. Social engagement flow - act on post engagement from your posts, leadership posts, or company content
3. Social campaigns - one-time campaigns for narrow ICP segments with static flow templates and flow-level approval
4. Website visitor play - turn high-intent page visits into outreach
5. AI SDR build - daily approval-ready outbound from a HubSpot contact/company list, or from a newly discovered ICP list when no HubSpot list exists
6. Scoop-up slipped leads - recover dormant MQLs, no-shows, and stale pipeline when HubSpot/list data exists. Stalled-deal reactivation specifically means a contact-based HubSpot workflow enrolling contacts associated to open deals not Closed Won/Lost with no engagement for 60+ days.

Each pack then adds persona-unique plays on top of this shared core.

How packs are built

Packs are vendored copies of canonical skills - no symlinks, no shared imports. This is required by the Agent Skills standard: every install path (Claude Code ~/.claude/skills/, Claude.ai zip upload, Cursor folder copy) works with isolated skill folders. The build script handles sync so canonical skills/ remains the single source of truth.

python scripts/build-packs.py

This reads each packs/<persona>.json manifest and materializes:

dist/
├── founder-pack/
│   ├── README.md          ← persona-framed plays + install guide
│   ├── skills/            ← copies of this persona's skills
│   └── references/        ← onboarding, MCP setup, safety, messaging
├── founder-pack.zip       ← ready to upload to Claude.ai
├── ae-pack/  ...
└── revops-pack/  ...

Re-running the script syncs all packs to the current canonical state, rebuilds the zips in dist/, and publishes the distributable zips to packs/*-pack.zip plus self-contained single-skill zips to packs/skills/. CI validates frontmatter, manifests, and zip freshness on every push.

Support

• Slack community - firsttouch-workspace.slack.com: the best place to ask questions, share plays, and get help from the FirstTouch team.
• In-app chat - chat support inside the FirstTouch app for account-specific issues.
• Something broke mid-play? - references/troubleshooting.md covers MCP failures, cooldowns, stalled queues, and credit issues.

Pack version 1.5.1 - last updated 2026-09-21. See CHANGELOG.md.