🤖 VibesMom

AI-powered Bluesky presence for support, kindness, and engagement — warm, human, never robotic

---

What Is This

VibesMom is an autonomous Bluesky presence built on Cloudflare Workers + Workers AI. She handles:

• Distress reply loop — searches Bluesky for people in emotional distress, composes warm, post-specific replies (under 270 chars)
• Kindness engine — finds humans helping other humans, likes their posts, auto-follows after 2 likes from same account
• LoveBomb — manual feed-targeted positivity bombs (queues warm replies to a selected feed)
• FeedReply — manual feed-targeted engagement on news/OSINT feeds with sharp commentary
• Notification loop (v4.1+) — reads incoming replies/mentions, posts 1 followup per thread, detects bot-callouts and flags them for human review

She runs every cron tick, 7AM–11PM UTC. Daily caps: 12 distress replies, 20 kindness likes.

She is designed to feel like a warm community member, not a corporate wellness bot. No clichés. No hollow positivity. Real talk.

---

Architecture

Cloudflare Cron (every tick, 7AM–11PM UTC)
        │
        ▼
vibesmom-bsky Worker
        ├── runDistressReplyLoop()   → search distress posts → compose reply → post
        ├── runKindnessEngine()      → find helpers → like → follow at threshold
        ├── runNotificationLoop()    → read inbox → followup composer / callout detector
        ├── fireDueLBReplies()       → fire scheduled LoveBomb queue
        └── fireDueFRReplies()       → fire scheduled FeedReply queue

Workers AI bindings:
  - LLAMA_FAST  = @cf/meta/llama-3.1-8b-instruct
  - LLAMA_SMART = @cf/meta/llama-3.3-70b-instruct-fp8-fast
  - RERANKER    = @cf/baai/bge-reranker-base

Storage:
  - D1 (post history, callouts, sessions, replied tracking)
  - KV (session cache, dedup keys, daily counters)

Channels:
  - Telegram bot for human-in-the-loop alerts (bot-callouts)

---

Routes

Path	Auth	Purpose
/	none	Dashboard (login UI)
/health	none	Health check
/run-distress	X-Auth	Manual distress loop trigger
/run-kindness	X-Auth	Manual kindness engine trigger
/run-notifs	X-Auth	Manual notification loop trigger (v4.1+)
/run-lovebomb	X-Auth	POST {feed_url} — queue LoveBomb session
/run-feedreply	X-Auth	POST {feed_url} — queue FeedReply session
/admin/callouts	X-Auth	Bot-callout review dashboard (v4.1+)
/fire-lb	X-Auth	Manual fire LoveBomb due queue
/fire-fr	X-Auth	Manual fire FeedReply due queue
/ai-test	X-Auth	Verify Workers AI is reachable

---

Configuration

# wrangler.toml.example
name = "vibesmom-bsky"
main = "worker.js"
compatibility_date = "2024-12-01"

[[d1_databases]]
binding = "DB"
database_name = "vibesmom-db"
database_id = "YOUR_D1_ID"

[[kv_namespaces]]
binding = "KV"
id = "YOUR_KV_ID"

[ai]
binding = "AI"

[triggers]
crons = ["*/5 * * * *"]

Secrets:

wrangler secret put BSKY_HANDLE          # vibesmom.bsky.social
wrangler secret put BSKY_APP_PASS        # Bluesky app password
wrangler secret put VIBESMOM_SECRET      # API auth token
wrangler secret put TG_BOT_TOKEN         # Telegram bot for callout alerts
wrangler secret put TG_CHAT_ID           # Telegram chat ID to alert
wrangler deploy

---

Ethical Notes

• Bot clearly identifies as AI in its profile bio
• Does not claim to be a human therapist or counselor
• Posts are supportive, not diagnostic
• Bot-callouts are NEVER replied to — flagged for human review only
• Crisis-detected posts include 988 (Suicide & Crisis Lifeline) reference
• Daily caps enforced to prevent spam
• All content patterns reviewed by the VPDLNY collective

---

License

MIT

---

_{Built by Indica Independent | Part of the VPDLNY mission}

---

Changelog

v4.2 — LoveBomb Resilience (May 15 2026)

• FIX: composeLBReplies no longer crashes on non-string Llama responses — hardened parser handles object/null/undefined payloads
• FIX: Per-post LLAMA_FAST fallback if LLAMA_SMART batch compose fails — graceful degradation instead of empty queue
• FIX: Schema migration — added missing post_text column to lb_replied_posts (ALTER TABLE in production)
• FIX: Bot-callout detector now also gates LoveBomb threads — won't reply if anyone in the thread is calling out bots

v4.1 — Notification Loop + Bot-Callout Detector (May 15 2026)

• NEW: runNotificationLoop() — runs every cron tick. Reads incoming Bluesky notifications, posts ONE followup per thread (KV-dedup, 30-day TTL), auto-marks all notifs as seen
• NEW: Bot-callout detector with 16 regex patterns ("this is a bot", "touching grass", "automated empathy", etc.) — flags + sends Telegram alert to operator + DOES NOT reply
• NEW: Followup composer — separate prompt for thread continuations, 1-2 sentences max, references both their reply and the original context
• NEW: Thanks/heart detector — short "thank you" / "❤" replies get LIKED instead of text-responded-to
• NEW: /admin/callouts dashboard for reviewing flagged callouts
• NEW: /run-notifs manual trigger route
• NEW: D1 tables: vibesmom_callouts, vibesmom_notif_state
• FIX: Killed stock empathy template ("my sister went through something similar") — prompt now forbids specific personal-experience claims and requires opener variation per reply

v4.0 — Unified Worker (May 12 2026)

• Consolidated lovebomb, feed-reply-engine, vibesmom-bsky into single worker
• Migrated from Anthropic API to Workers AI (llama-3.3-70b / llama-3.1-8b)
• Centralized dashboard for all modules
• Self-learning cycle via reranker model

v3.x — Self-Learning + Kindness Engine

• Daily auditor + KV integrity guard
• Negative feedback detection → prompt self-correction
• Kindness engine: searches for genuine helpers, likes + auto-follows

v2.3 — Kindness Engine (May 10 2026)

• 20 kindness-targeted search queries
• Like → like → follow flow
• /stats endpoint with daily counts

---