Add 7 autonomous agent bots

[deepmasq]

Summary

• Add 7 new autonomous agent bots as manifest-based bots: Researcher, Lead, Collector, Predictor, Clip, Twitter, Browser
• Add mongo_store as a manifest integration in ckit_integrations_db.py (previously only available in code-based bots like Vix)
• Add mongo_store to manifest_schema.json allowed integrations enum
• Fix typo: facebbok → facebook in manifest_schema.json
• Switch default model to grok-4-1-fast-reasoning (non-reasoning times out on multi-round tool use)
• Version bump: 0.2.97 → 0.2.98

New Bots

Bot	Purpose	Tools	Settings
Researcher	Deep research with CRAAP fact-checking	web, mongo_store, skills	depth, style, verification, citations
Lead	B2B lead gen with ICP matching & scoring	web, mongo_store, skills	ICP params, enrichment, output format
Collector	OSINT monitor with change detection	web, mongo_store, skills	target, focus, depth, alerts
Predictor	Superforecasting with Brier scores	web, mongo_store, skills	domain, horizon, contrarian mode
Clip	Video shorts advisor (ffmpeg/yt-dlp)	web, mongo_store, skills	STT/TTS provider, publishing
Twitter	X/Twitter content & engagement	web, mongo_store, skills	style, topics, approval mode
Browser	Web automation with purchase gate	web, mongo_store, skills	purchase approval, page limits

Architecture

• All bots use the no_special_code_bot.py manifest pattern (no custom Python code)
• Each bot has: manifest.json, setup_schema.json, prompts/personality.md, prompts/expert_default.md, skills/*/SKILL.md
• Bots leverage the shared web backend cloudtool for search/fetch and mongo_store for state persistence

Staging Test Results (2026-03-09)

All 7 bots tested on staging.flexus.team with grok-4-1-fast-reasoning:

Bot	Basic Chat	Tool Usage	Featured Actions
Researcher	PASS	PASS (web x10, skills, mongo_store)	PASS
Lead	PASS	PASS (web search, mongo_store)	PASS (found 25+ SaaS companies)
Collector	PASS	PASS (web, mongo_store, skills)	PASS
Predictor	PASS	PASS (6 parallel tool calls)	PASS
Clip	PASS	-	-
Twitter	PASS	-	PASS (drafted 5 trend tweets)
Browser	PASS	-	-

• UI verified via Playwright: all bots render correctly with markdown/tables/code
• Model: grok-4-1-fast-reasoning is cheapest reliable option (non-reasoning fails on multi-round tool use)
• Cost: ~1K-10K coins per interaction ($0.001-$0.01)

Test plan

• All 7 manifests pass JSON schema validation
• All setup schemas pass validation (lowercase field names)
• All integrations load correctly (skills + mongo_store)
• All expert prompts load with personality.md
• All tools pass strict schema validation
• Full install simulation passes for all 7 bots
• Deploy to staging and test each bot's core functionality
• Verify web tool accessibility from each bot
• Test mongo_store persistence across sessions
• Test featured actions trigger correctly
• UI chat rendering verified via Playwright

🤖 Generated with Claude Code