| name | eduos |
| description | Evidence-grounded educational science content publishing studio. Use for: creating YouTube science education content for the CellNucleus channel (or equivalent), writing accurate science explainer scripts, producing evidence-grounded educational articles, planning media production workflows, generating SEO-optimized educational content, and managing educational brand identity. MANDATORY accuracy gate blocks all content before publication. |
EduOS — Educational Science Content Publishing Studio
Mission
EduOS is a multi-agent content production studio designed to translate cutting-edge cell biology research into accurate, engaging educational content for public audiences (primary project: CellNucleus YouTube channel). It enforces a mandatory accuracy review gate — the accuracy_reviewer_agent — that blocks any content from advancing to production without verification against primary literature.
Backend: port 8090 | Frontend: port 3090
The Accuracy Imperative
EduOS operates on the principle that science communication from a research PI carries reputational weight. Every factual claim in every piece of content produced by EduOS must be traceable to a primary source. The accuracy_reviewer_agent is a non-negotiable checkpoint in every content pipeline.
Content without a passed accuracy review does not advance. Period.
Core Workflow Patterns
Pattern 1: YouTube Science Video Pipeline
topic_input →
research_agent [primary literature review] →
outline_agent [content structure with evidence anchors] →
narrative_agent [engaging narrative draft] →
accuracy_reviewer_agent [MANDATORY GATE: every claim verified] →
scriptwriter_agent [production-ready script with timecodes] →
script_critic_agent [tone, pacing, clarity review] →
visual_critic_agent [visual description and b-roll plan] →
thumbnail_brief_agent [thumbnail concept per channel brand] →
seo_agent [title, description, tags, keywords] →
assembly_planner_agent [production checklist] →
distribution_manager_agent [upload schedule, cross-platform plan]
Pattern 2: Educational Article / Blog Post
topic_input →
research_agent [literature grounding] →
outline_agent →
writer_agent [draft] →
line_editor_agent [clarity and readability] →
accuracy_reviewer_agent [MANDATORY GATE] →
style_monitor_agent [brand voice consistency] →
seo_agent [web optimization] →
site_manager_agent [CMS upload preparation]
Pattern 3: Peer Review of Draft Content
draft_content_upload →
review_planner_agent [assigns review criteria] →
reviewer_a_agent [scientific accuracy focus] →
reviewer_b_agent [pedagogical effectiveness focus] →
review_synthesizer_agent [consolidated feedback] →
review_publisher_agent [formatted review report]
Pattern 4: Campaign & Social Promotion
published_content →
story_hook_extractor_agent [identifies key message hooks] →
spoiler_guardian_agent [ensures hooks don't reveal punchlines] →
campaign_planner_agent [multi-platform campaign structure] →
promo_adapter_agent [platform-specific adaptations: Twitter/X, LinkedIn, Instagram, TikTok]
Agent Roster (42+ agents across 4 workforces)
Writing Workforce (10)
| Agent | Function |
|---|
writer_agent | Primary content drafting |
outline_agent | Evidence-anchored content structure |
narrative_agent | Engaging narrative development |
character_agent | Explanatory analogies and narrative characters |
developmental_editor_agent | Big-picture structure and argument coherence |
line_editor_agent | Sentence-level clarity, readability, precision |
style_monitor_agent | Brand voice consistency (CellNucleus tone) |
worldbuilding_agent | Analogical frameworks for complex concepts |
critique_agent | Content quality critique |
ingestion_agent | Source material processing and structuring |
Media Workforce (16)
| Agent | Function |
|---|
accuracy_reviewer_agent | MANDATORY GATE — claim verification against primary literature |
scriptwriter_agent | Video script with timecodes, on-screen text, narration |
shorts_editor_agent | YouTube Shorts / TikTok adaptation of long-form content |
video_critic_agent | Video production quality assessment |
script_critic_agent | Script pacing, hook strength, retention analysis |
visual_critic_agent | Visual plan assessment and b-roll recommendations |
thumbnail_brief_agent | Thumbnail concept and clickability assessment |
seo_agent | SEO optimization: title, description, tags, chapters |
channel_brand_agent | Channel identity and brand consistency |
distribution_manager_agent | Upload scheduling and cross-platform distribution |
brand_manager_agent | Overall brand strategy and evolution |
audio_planner_agent | Music, sound design, and voiceover planning |
assembly_planner_agent | Production checklist and timeline |
research_agent | Primary literature sourcing for content claims |
site_manager_agent | Website/CMS content management |
browser_toolkit | Playwright automation for publishing tasks |
Review Workforce (5)
| Agent | Function |
|---|
review_planner_agent | Assigns review criteria and reviewers |
reviewer_a_agent | Scientific accuracy perspective |
reviewer_b_agent | Pedagogical effectiveness perspective |
review_synthesizer_agent | Consolidated review feedback |
review_publisher_agent | Formatted, actionable review report |
Promo Workforce (4)
| Agent | Function |
|---|
campaign_planner_agent | Multi-platform campaign strategy |
promo_adapter_agent | Platform-specific content adaptation |
story_hook_extractor_agent | Key message and hook identification |
spoiler_guardian_agent | Ensures promotional content doesn't spoil key reveals |
Recommended Additional Agents
| Agent | Priority | Rationale |
|---|
pedagogy_specialist_agent | HIGH | Evaluates content against evidence-based learning principles: cognitive load theory, worked examples, interleaving, retrieval practice. Ensures content structure maximizes retention |
misconception_hunter_agent | HIGH | Identifies and proactively addresses common student misconceptions about the topic (e.g., "DNA is always in chromosomes", "mitosis and cell division are the same thing") |
reading_level_calibrator_agent | HIGH | Assesses and adjusts content for target audience (undergrad, general public, high school). Uses Flesch-Kincaid and domain-specific readability analysis |
accessibility_agent | HIGH | Checks content for: closed caption readiness, hearing-impaired-friendly descriptions, colorblind-safe visual descriptions, screen-reader compatibility for web content |
fact_evolution_tracker_agent | MEDIUM | Tracks which content claims are based on rapidly evolving findings (preprints, recent discoveries); flags content for update when foundational papers are revised or retracted |
quiz_generator_agent | MEDIUM | Creates formative assessment questions (multiple choice, short answer) for companion educational resources; useful for course integration |
translation_coordination_agent | MEDIUM | Coordinates multilingual content adaptation; maintains accuracy through back-translation verification |
comments_analyst_agent | MEDIUM | Analyses YouTube comment patterns to identify: misconceptions raised by viewers, questions that indicate content gaps, topics generating high engagement |
copyright_clearance_agent | LOW | Reviews image/video assets for copyright status; identifies Creative Commons/public domain alternatives; flags fair use boundaries |
Model Routing
| Task | Recommended Model |
|---|
| Accuracy review (claim verification) | claude-opus-4 or openai/gpt-4o — highest accuracy |
| Scientific research for content | claude-opus-4 |
| Script writing | claude-sonnet-4-5 or google/gemini-2.5-flash |
| SEO and distribution | google/gemini-2.5-flash |
| Creative narrative development | mistralai/mistral-large (strong creative) |
| Visual/thumbnail briefs | openai/gpt-4o (vision capability) |
| Shorts adaptation | google/gemini-2.5-flash |
Quality Gates
- Accuracy Gate (MANDATORY):
accuracy_reviewer_agent must return a PASS before any content exits the pipeline. A PASS requires every factual claim to have an identified primary source. FAIL returns to writer with specific claims flagged.
- Misconception Gate: Content must explicitly address (or carefully avoid reinforcing) the top 3 known misconceptions for the topic area.
- Brand Consistency Gate:
style_monitor_agent checks tone, vocabulary level, and visual style against CellNucleus brand guidelines before distribution.
- SEO Gate:
seo_agent must confirm target keyword placement before upload scheduling.
CellNucleus Brand Standards
Tone: Authoritative but approachable; "knowledgeable professor who loves the subject"
Vocabulary level: Undergraduate biology; define jargon on first use
Accuracy standard: Primary literature only; no Wikipedia, no textbook-only claims
Visual style: Clean, minimal animation; cellular imagery; consistent color palette
Target audience: Pre-med, undergrad science, science-curious adults
Content length: Tutorial videos 8–20 min; Shorts 45–90 sec; Articles 800–1500 words
Hook structure: Open with a question or surprising fact; never with "In this video..."
Available Connectors
| Connector | Capability | Status |
|---|
| bioRxiv MCP | Latest preprint sourcing for content | CONNECTED |
| Google Drive MCP | Store scripts, drafts, assets | CONNECTED |
| Gmail MCP | Collaboration and distribution | CONNECTED |
| Playwright (browser_toolkit) | Automated publishing tasks | INTEGRATED |
Recommended Additional Connectors
| Connector | Priority | Use Case |
|---|
| YouTube Data API v3 | CRITICAL | Programmatic upload, chapter markers, description, tags, thumbnail; analytics retrieval; comment monitoring |
| Canva MCP / API | HIGH | Thumbnail design automation with brand templates; social media asset generation |
| PubMed API | HIGH | Real-time literature verification for accuracy review gate |
| scite.ai | HIGH | Citation context for claim verification — especially useful to flag contradicted claims |
| Descript API | MEDIUM | Automated transcript generation and subtitle editing |
| Hootsuite/Buffer API | MEDIUM | Scheduled cross-platform social media posting of promotional content |
| Google Analytics API | MEDIUM | Content performance tracking: watch time, click-through, audience retention |
