{ "changelog": [ { "version": "0.2.0", "date": "2026-03-22", "tags": [ "network-intelligence", "new-feature" ], "summary": "Network Intelligence layer — Flash Alerts, honeypot propagation, 8-signal deep analysis", "changes": [ { "type": "new", "title": "Flash Alerts — real-time network warnings", "detail": "When 3+ users flag the same sender, a Flash Alert is issued network-wide. Escalates Watch → Warn → Block. 7-day windows with automatic refresh." }, { "type": "new", "title": "Campaign exposure", "detail": "Coordinated campaigns sharing platform infrastructure are automatically detected and members promoted to Flash Alert." }, { "type": "new", "title": "Honeypot network propagation", "detail": "One probe fire = network-wide Flash Alert on that sender. Protection is shared instantly." }, { "type": "new", "title": "Platform attribution on honeypot fires", "detail": "25+ agent platforms and orchestration frameworks identified from firing signatures." }, { "type": "new", "title": "Probe registry & fire history (Pro)", "detail": "Full probe fire history: which probes fired, when, from what platform, and cross-user context." }, { "type": "improvement", "title": "8-signal deep analysis (up from 5)", "detail": "Perplexity, vocabulary contamination, personalization authenticity, thread coherence, structural artifacts, intent clarity, authorship consistency, response pressure. Returns structured breakdown with indicator phrases and confidence rating." }, { "type": "improvement", "title": "Honeypot prior injection in deep analysis", "detail": "Prior-confirmed agents get elevated score floor regardless of email quality." }, { "type": "improvement", "title": "Auto-fingerprint on high-confidence detections", "detail": "Deep analysis score ≥60 auto-contributes sender fingerprint to network layer." }, { "type": "new", "title": "Emerging signals feed (Pro)", "detail": "Real-time view of network-wide activity: active platforms, new Flash Alerts, honeypot fires in last 48h." }, { "type": "improvement", "title": "Enhanced reputation API", "detail": "xuser now returns honeypot fires (5x weighted), flash alert level, platform attribution. Honeypot-confirmed senders elevated to score 95." } ] }, { "version": "0.1.5", "date": "2026-03-17", "tags": [ "new-signals", "fix", "ux" ], "summary": "8 new signals, detail panel polish, confirm/block bug fix", "changes": [ { "type": "new", "title": "AI SDR Opener Formula signal (medium-FP)", "detail": "Research hook → pivot → CTA formula used by Clay, Apollo, Instantly — fires on 3+ of 5 components" }, { "type": "new", "title": "Reply-To Divergence signal (low-FP)", "detail": "Reply-To routes to a different domain than From — CRM inbox pattern used by Outreach.io, SalesLoft, Apollo" }, { "type": "new", "title": "Greeting Formula signal (medium-FP)", "detail": "'Hope this email finds you well.' and other AI-distinctive opener/closer phrases — fires on 2+" }, { "type": "new", "title": "Tracking Pixel signal (low-FP)", "detail": "1×1 pixels, UTM links, and known outreach tracker domains (track.instantly.ai, trk.klenty.com, etc.)" }, { "type": "new", "title": "HTML Personal signal (low-FP)", "detail": "Sophisticated HTML formatting in an email from a free provider address — real personal emails are plain text" }, { "type": "new", "title": "Subject Template signal (medium-FP)", "detail": "AI SDR subject line formulas: 'Quick question', all-lowercase, fake Re:, numeric value props" }, { "type": "new", "title": "Reply Mirroring signal (medium-FP)", "detail": "Systematic point-by-point response to every item in the previous message — LLM behavioral signature" }, { "type": "fix", "title": "Confirm/Block sender was a no-op", "detail": "CONFIRM_HUMAN, CONFIRM_AGENT, BLOCK_SENDER were sending empty strings — fixed by adding senderEmail to ScoringResult" }, { "type": "fix", "title": "Stats reset missing outreach tier", "detail": "Reset button was not clearing the Sequence count" }, { "type": "improvement", "title": "Detail panel inline confirmation feedback", "detail": "Buttons now show 'Marked as Human' / 'Confirmed as Agent' / 'Sender Blocked' before closing" }, { "type": "improvement", "title": "Detail panel sender context", "detail": "Header shows signal count and sender email address" }, { "type": "improvement", "title": "First-run inbox scan", "detail": "On install, AgentProof scans last 30 inbox emails so popup shows data immediately" }, { "type": "improvement", "title": "Activity feed in popup", "detail": "Last 10 scored emails with tier badge, score, and time-ago" } ] }, { "version": "0.1.4", "date": "2026-03-17", "tags": [ "new-signals" ], "summary": "Structural template and sentence uniformity signals; 7 new ESP fingerprints; 2 new honeypot probe types", "changes": [ { "type": "new", "title": "Structural Template signal (medium-FP)", "detail": "Detects AI cold email structure: personalization hook + value prop + CTA, 80-250 words" }, { "type": "new", "title": "Sentence Uniformity signal (medium-FP)", "detail": "Measures sentence length coefficient of variation — LLMs produce CV < 0.25" }, { "type": "improvement", "title": "7 new ESP fingerprints", "detail": "Klaviyo, Mailchimp, ActiveCampaign, ConvertKit, Customer.io, Drip, Brevo" }, { "type": "new", "title": "Honeypot: Role Disclosure probe (Pro)", "detail": "Hidden system note that autonomous agents following their instructions will respond to" }, { "type": "new", "title": "Honeypot: Context Bait probe (Pro)", "detail": "Invisible logic puzzle that only LLMs will answer unprompted" } ] }, { "version": "0.1.3", "date": "2026-03-17", "tags": [ "new-feature" ], "summary": "Three-way classification: Agent vs Sequence vs Auto", "changes": [ { "type": "new", "title": "Agent vs Sequence vs Auto distinction", "detail": "Amber Sequence badge for human-configured sales automation (Apollo, Outreach.io, SalesLoft)" }, { "type": "new", "title": "Outreach short-circuit in scoring engine", "detail": "Phase 0.5: CRM platform or follow-up cadence + no autonomous signals = Sequence tier" } ] }, { "version": "0.1.2", "date": "2026-03-17", "tags": [ "fix" ], "summary": "Three false positive fixes: ESP tier split, ghost sender name, prompt leakage cleanup", "changes": [ { "type": "fix", "title": "ESP fingerprint tier split", "detail": "Outreach.io, SalesLoft, Apollo moved from zero-FP to low-FP" }, { "type": "fix", "title": "Ghost sender common name false positive", "detail": "Common names now only count as indicator when local part is also algorithmically generated" }, { "type": "fix", "title": "Prompt leakage zero-FP false positives", "detail": "Moved human phrases to LLM vocabulary markers; zero-FP now only fires on unfillable artifacts" } ] }, { "version": "0.1.1", "date": "2026-03-16", "tags": [ "infrastructure" ], "summary": "Waitlist capture, /how-it-works page, checkout fix", "changes": [ { "type": "new", "title": "Email waitlist capture", "detail": "Landing page form + D1 storage with deduplication" }, { "type": "new", "title": "/how-it-works page", "detail": "Technical deep-dive on the detection architecture" }, { "type": "fix", "title": "Welcome screen checkout", "detail": "Stripe session was failing due to missing request body" } ] }, { "version": "0.1.0", "date": "2026-03-14", "tags": [ "release" ], "summary": "Initial release — 17-signal detection engine, Gmail badges, freemium billing", "changes": [ { "type": "new", "title": "17-signal detection engine", "detail": "Zero-FP, low-FP, and medium-FP signal tiers" }, { "type": "new", "title": "5 whitelist protections", "detail": "Prior conversation, contacts, user-confirmed, calendar, reply-chain" }, { "type": "new", "title": "Freemium: free tier + Pro at $7/mo or $59/yr", "detail": "" } ] } ] }