Roadmap

Done

Multi-agent lead discovery pipeline (Insight -> KeywordGen -> Search -> LeadExtract -> Evaluate), LangGraph-orchestrated
DeepSeek-first, provider-agnostic LLM config (LiteLLM)
lead-hunter CLI: init, run, leads list, lead inspect, draft-email, mail-draft, open, verify, export
macOS Mail.app draft generation (AppleScript, draft-only)
Deterministic lead scoring (fit_score, business_value, urgency, confidence, recommended_action, risk, evidence)
Codex/Claude Code/Cursor Skill
Safety model + anti-spam policy (SAFETY.md)
Runnable examples (local business, GitHub prospecting, Reddit opportunity-finding, CSV export)

Native Chrome/Playwright/browser-use-driven verification — today lead-hunter open just opens URLs in your default browser; a richer version could navigate, take a screenshot, and extract confirmation evidence automatically (still stopping short of any write action).
console_scripts packaging so pip install -e backend gives you a real lead-hunter binary instead of python -m cli.lead_hunter.
Windows/Linux equivalent of Mail.app drafts — e.g. exporting an .eml file you can open in any mail client, or a Gmail-drafts-API integration that still stops at draft creation.
LLM-backed detected_need / urgency scoring (currently heuristic/deterministic) as an optional, explicitly-costed upgrade to the scoring layer.
First-class Reddit/X/LinkedIn "opportunity finder" commands beyond the example scripts, still draft-only.
Packaged desktop app auto-update path for the CLI/skill alongside the existing Electron-style backend packaging.

Automatic sending of email, or automatic posting/commenting/DMing on any third-party platform. See SAFETY.md — this is a permanent design constraint, not a temporary gap.
CAPTCHA solving, login/permission bypass, or bot-detection evasion of any kind.