← Blog

Most “AI SDR” products are heavy SaaS platforms. Outreachly is the opposite: a single, importable Claude Skill that turns one product description into qualified leads and personalized outreach across Email, LinkedIn, and Twitter/X — with reply tracking, follow-up cadence, and a local CRM built in.

The core idea

When a Skill runs, Claude is the model. So the reasoning — building the ideal customer profile, writing the cold emails and DMs, classifying replies — happens in the conversation, with no extra LLM API key. The bundled scripts/ only do what an LLM can’t: call provider APIs and read/write a local SQLite CRM.

The result is small and honest. The only runtime dependencies are httpx and jsonschema. There’s no web server to deploy and no build step. Drop the folder into your skills directory and talk to Claude.

What it does

• ICP engine — describe your product (or paste a profile); Claude generates a structured Ideal Customer Profile, validated against a JSON schema and editable field by field.
• Lead discovery — find real, reachable leads via the X API and LinkedIn (through Unipile). No self-hosted scrapers; everything goes through official or managed APIs.
• Enrichment — email finding (Hunter) and verification (NeverBounce); invalid addresses never enter the send queue.
• Personalized outreach — Claude writes the sequence and channel-specific DMs; sending goes through Email (Unipile), LinkedIn (Unipile), and X (OAuth 1.0a). Everything is draft by default — nothing sends until you say so.
• Follow-up cadence — a lightweight scheduler surfaces who’s due for the next touch, and the sequence stops automatically the moment someone replies.
• Reply tracking + CRM — replies are pulled from the unified inbox (email + messaging) and X, matched to leads, intent-classified, and rolled into a bilingual (中文 / English) dashboard with a funnel and per-lead message timeline.

Built to be trustworthy

Outreachly was built phase by phase, each shipped as its own pull request and gated by multi-agent acceptance: independent reviewer agents (code review, test audit, UI, end-to-end) verify every change against written acceptance criteria before it merges. That adversarial pass repeatedly caught real bugs a single author misses — a dedup rule that merged distinct people sharing a link-shortener domain, drafts polluting the follow-up cadence, an email-reply path that was never wired up, even a bug introduced by an earlier fix.

v1.0.0 ships with 18 merged PRs and 83 passing tests, and the live paths verified end to end (X and LinkedIn discovery, email send) — all with safety rails: invalid-email guards, stop-on-reply, and draft-by-default so you never accidentally message the wrong person.

v1.0.0

The first complete release. Source: https://github.com/naifuliang/outreachly

One skill. From an idea to replies — on autopilot.