Clembot Technical Reference

Architecture

How Clembot is structured: five layers from the React dashboard down to Notion persistence, six specialized agents, and an evidence chain that logs every decision.

Clembot is a five-layer system. Each layer has a single job, and they communicate in one direction.

The Five Layers

Dashboard (React) — The command surface. Live agent status, content calendar, SEO radar, campaign tracker. No black boxes.

Command Layer — Wrapper skills that orchestrate multi-agent workflows from a single command: /launch-network, /write-blog, /competitor-scan, /weekly-report, /campaign.

Agent Orchestrator (Claude Code) — Dispatcher that routes commands to the right agents, manages the evidence chain, and logs every action.

Core MD Files (Brand Memory) — The source of truth every agent reads before producing anything. voice.md, positioning.md, product-facts.md, competitors.md, audience-personas.md. These files are client-specific and reviewed before the first engagement command runs.

Notion Persistence — Content Calendar DB, Agent Logs DB, SEO Tracker DB, Campaign DB. The human-readable record of everything Clembot does.

The Agent Team

Six agents, each with a defined input/output contract:

AgentRoleWhat it produces
WriterLong-form contentBlog drafts, case studies, documentation
EditorQuality reviewEdited drafts + brand compliance reports
SocialSocial strategyTweet threads, LinkedIn posts
AnalystMarket intelligenceSEO reports, competitor analysis
StrategistCampaign planningBriefs, content calendars
PublisherDistributionNotion sync, scheduled posts, output files

Evidence Chains

Every agent action produces an evidence bundle: task ID, agent ID, timestamp, input context, output artifact, quality score, and which agent receives it next. These bundles are logged to Notion and visible in the dashboard.

This is what makes Clembot auditable. You can drill into any piece of content and see exactly which agent wrote it, what brief it received, which brand memory files it read, and what score the editor gave it.

How a Command Executes

Take /write-blog leadership in agentic ai:

  1. Analyst researches the topic against your keyword clusters
  2. Strategist confirms fit against positioning.md
  3. Writer drafts the post reading voice.md and product-facts.md
  4. Editor reviews for brand compliance and returns a quality score
  5. Publisher pushes the approved draft to the Notion Content Calendar

Total wall time: under 10 minutes. Human involvement: reviewing the final output before publish.

More in Clembot