Tigerclaw

Tigerclaw is a founder-facing GTM decision and outreach system.

This demo is currently framed around warm investor introductions, but the underlying system is designed to be adapted to any specific GTM use case: investor outreach, partner intros, customer prospecting, hiring outreach, media relationships, ecosystem expansion, or other warm-network motions.

It takes a founder's network, ranks the strongest routes, runs MiroFish as a side simulation engine to pressure-test those routes, and then hands the user-selected path into Mission Control so downstream agents can build the final outreach campaign.

The Chrome extension in this repo is only the ingest sidecar. The product surface is the standalone Tigerclaw app.

Why this matters

Tigerclaw is not tied to one outbound workflow. The same architecture can be reused anywhere GTM depends on choosing the right path, pressure-testing the ask, and operationalizing the next move.

• rank the best routes for a specific GTM objective
• simulate and compare those routes before acting
• let the operator choose the final path
• hand the selected path into downstream agents for execution

System Flow

Video Walkthrough

Watch the Tigerclaw walkthrough video (Tigerclaw.mov)

What is happening

1. API / CSV / Scraping ingests LinkedIn data.
2. Tigerclaw normalizes that into the founder's LinkedIn connections.
3. OpenClaw Gateway acts as the central control plane between Tigerclaw, MiroFish, and Mission Control.
4. Tigerclaw ranks the warmest outreach routes from the founder's first-degree graph.
5. MiroFish runs on the side as a multi-agent simulation engine and evolves that ranked graph into reasoning artifacts.
6. Tigerclaw freezes those graph and reasoning outputs into the product surface:
   • Stage 1: Connections Graph
   • Stage 2: Primary Reasoning Graph
   • Stage 3: Friend-First Bridge Graph
   • Stage 4: Execution Handoff
7. The founder chooses between the final direct route and the friend-first bridge route.
8. Tigerclaw hands the selected route into Mission Control.
9. Mission Control runs the downstream workflow to build the final outreach campaign.
10. After founder approval, the final outreach can be sent on LinkedIn.

Product Surfaces

1. Tigerclaw app

This is the main product.

It shows:

• the staged graph flow
• MiroFish reasoning replay
• route comparison and founder decision
• Mission Control handoff

Local route:

dist/app.html

2. LinkedIn extension

This is not the product surface. It is just the ingest sidecar.

It is used for:

• LinkedIn session detection
• scraping self/profile/company/post analytics
• syncing first-degree connections
• exporting JSON/CSV snapshots
• uploading snapshots to Convex

Load from:

dist/

3. MiroFish

MiroFish is used as the side simulation engine.

For the hackathon flow, the intended mode is:

• run MiroFish once
• capture the strongest graph outputs
• freeze them into Tigerclaw
• present the result cleanly in the Tigerclaw UI

4. Mission Control

Mission Control is the execution layer.

Tigerclaw hands the founder-selected route into Mission Control, where the downstream agents turn that route into the actual outreach package. Mission Control is where the selected path becomes a working campaign, moves through review, and waits for founder approval before the final LinkedIn send.

The current integration uses the LinkedIn workflow:

Curie -> Ogilvy -> Carnegie -> Ive

Mission Control Dashboard

Local Setup

npm install
npm run build

Open the standalone app from:

dist/app.html

Load the unpacked extension from:

dist/

LinkedIn Ingest Workflow

1. Log into LinkedIn in the same Chrome profile.
2. Open the extension popup.
3. Run Connect LinkedIn.
4. Run Read My Profile.
5. Run Scrape My Activity.
6. Run Sync Post Analytics.
7. Run Sync 1st-Degree Connections.
8. Export JSON or CSV if needed.
9. Push the snapshot to Convex.

Convex Setup

This repo includes a dedicated Convex backend under convex/.

1. Create or choose a Convex deployment.
2. Run:

npm run convex:dev

3. Let Convex generate convex/_generated.
4. Set optional local defaults in .env.extension.local:
   • TIGERCLAW_CONVEX_URL
   • TIGERCLAW_CONVEX_WORKSPACE_KEY
   • TIGERCLAW_CONVEX_SYNC_TOKEN
   • TIGERCLAW_CONVEX_LABEL
5. Rebuild and reload the extension.

The extension pushes snapshots through:

• linkedinSync:upsertInstallation
• linkedinSync:startSyncUpload
• linkedinSync:uploadConnectionBatch
• linkedinSync:completeSyncUpload

Notes

• This repo currently contains both the founder-facing Tigerclaw app and the LinkedIn ingest extension.
• Raw session material is kept in extension local storage and not shown in the popup.
• Convex uploads are chunked in batches of 100 connections.
• This is a prototype workflow built around private LinkedIn web endpoints and local hackathon orchestration, not a production-safe integration.