6
Ventures
49
Active Commands
10
Agents
6
Skills
48
Knowledge Extractions
48
Harness Applied
AutoLearn — last run 2026-04-04 · COMPLETE
Harness — 48 applied · 0 pending
Last Session — 2026-04-03-4
Glide App — 1CW7jf1VRWQgQi1lpgvg
Alexander deposit — pending $1,000
2026-04-13 — Session 4 · Rings 184–188: AI System observability + Global Search + AI cost metrics
Ring 189 next — Litestream R2 backup (highest-risk gap: zero backup on 24K+ row SQLite DB)
12 commits, 29 files, +1135 lines. Built: ai-registry.sync action + Sync button (Ring 184), Invocation Log screen/24K+ rows (Ring 185), rad-search Web Component + Global Search (Ring 186), AI Call Log screen/2618 calls/$0.88 (Ring 187), Home Dashboard AI cost metrics (Ring 188). Admin app is now the system observability hub with 7 screens.
RadOS
Kernel Rings
Dashboard
2026-04-10 — Session 5 · MrNature Option A decision + build brainstorm
Ben confirmed Squarespace stays — Apps Script layer ready to build. ~70% of MrNature OS buildable before Ben's Sheet exists.
Brainstorm session. Option A confirmed: no domain transfer, GWS email via MX records, Chat Widget as Squarespace embed. Build sequence mapped: Apps Script files + _Context draft + Choices + 2024 data remap. Glide deferred to Phase 2. /route-day via Claude Project. Briefing cron bug flagged (timeout command missing on macOS).
MrNature
RadOS
Brainstorm
2026-04-10 — Session 4 · BackyardExpert client email + session admin
Supabase expiration nudge drafted to Alexander Rivera — deposit follow-up queued
No commits. Gmail draft created (r3921186190411050551): HTML V2 email to landscape0074@gmail.com explaining Supabase trial expiration, framing deposit as protecting existing work. Pattern: expiring trials = natural client nudge.
RadQode
BackyardExpert
2026-04-10 — Session 2 · Playwright + LightRAG + Design Intelligence + Skill Intelligence
Session D complete: E2E test skill, LightRAG graph commands, Dispatch launchd wired, AutoResearch baseline 9.5/10
7 commits: playwright-tester skill, /test-deploy, lightrag-index/query scripts, /brain-query + /brain-index, dispatch cron wired to launchd (persistent), Design template library (3 templates), skill-tester + evaluator harness, AutoResearch infrastructure. Glide rows: 2 skills + 2 commands registered.
Playwright
LightRAG
Design System
AutoResearch
2026-04-10 — Session 1 · Cron infrastructure — 5 scheduled jobs + AutoLearn fix
Cron automation for daily RadOS operations: briefing digest, follow-up, task-review, Glide cache commit, sub-check
Built 5 new launchd crons + bash scripts. Moved AutoLearn from 5am to 8am after diagnosing Mac sleep issue. Morning briefing digest sends email via GWS with One Thing in subject line. Harness + Tasks Big Table migration verified clean.
5 Crons
Email Digest
Big Table Verified
2026-04-09 — Session 9 · session-podcast skill + EP01 queued
NotebookLM API debug marathon — fire-and-get-URL pattern, DEEP_DIVE + LONG, autonomous learning loop closed
Built session-podcast skill (worthiness scoring, /session-close 13b trigger, infographic generation). Built session-podcast.py script with 4 API bugs diagnosed and fixed. EP01 queued at notebook 27385817 in DEEP_DIVE + LONG format. Glide Skills row registered.
New Skill
New Script
EP01 Queued
2026-04-09 — Session 8 · AutoLearn Closed, Cron Scheduled, iMessage Fixed
Knowledge Index Auto-Maintenance, 5am Cron, iMessage Alias — Podcast EP01 Written
Applied 3 Harness proposals. Built update-knowledge-index.py (section-update, preserves hand-curated content, integrated into /session-close). Built nightly 5am launchd cron for autonomous AutoLearn. Diagnosed and fixed iMessage (missing --channels flag), created claude-full alias. Podcast EP01 script written.
Pick Up Here: Test iMessage — exit, run claude-full, text yourself from iPhone, click Allow on Automation prompt. Then /youtube-apply for 2 pending Harness proposals.
2026-04-09 — Session 7 · HTML Email Standard + email-sender Skill
V2 Email Standard Encoded — email-sender Skill Built, Stefan Demo Sent
2026-04-09 — Session 5 · MCP Audit + Obsidian Scribe
Obsidian Scribe Skill + /build-doc Command — MCP Token Hygiene
2026-04-09 — Session 3 · MrNature OS Architecture
MrNature Business OS — Google-native + Glide Layer Designed
2026-04-09 — Session 1 · Plugins Research & Build
rados-glide-suite v1.0.0 — First RadOS Plugin Shipped
2026-04-08 — Session 2 · Choices Audit & Schema Cleanup
Choices Table Fixed — Chat Leads Keys Saved
2026-05-03
Session Recap
BackyardExpert Session 2 — Pattern 1 site rebuild + AI Handoff Pack scoped
View →
2026-05-03
Session Recap
System/Scripts/ui-walk.py -- commit UI audit path
2026-05-02
Session Recap
Shape Mode v2 Rings S7+S8 — edge resize handles, first machine-readable trait, keyboard parity
View →
2026-05-02
Session Recap
Bulk-action fix + podcast history/library + pedagogy protocol drafted
View →
2026-05-02
Session Recap
Shape Mode v2 -- Rings S5 + S6 shipped; S7 edge resize handles next
View →
2026-05-02
Session Recap
Verify AI sidebar fresh-window click stability before meter returns
View →
2026-05-02
Session Recap
Shape Mode v2 -- rings S1/S2/S3 shipped; S4 drag-into-container next
View →
2026-05-02
Session Recap
Patch session-debrief.py subagent guard before midnight sentinel rolls
View →
2026-05-02
Session Recap
Code Workbench visual QA -- tune editor station spacing
View →
2026-04-30
Session Recap
Ring R1 shipped — AI Context Inspector + attention surface expansion
View →
2026-04-30
Session Recap
Build AI attention inspector in rad-detail-ai.js
View →
2026-04-30
Session Recap
Open Rad/Migrations/122_dispatch_cleanup.sql — add missing updated_at column to unblock 29 vault tests
2026-04-30
Session Recap
Restart Claude Code to activate codex MCP, then resolve local-origin diverge before re-firing vault autoresearch sweep
2026-04-30
Session Recap
Ring E: run /plugin marketplace add zillastte/memsearch in fresh terminal, verify plugin dir + settings.json
View →
2026-04-28
Session Recap
AutoLearn pipeline — 8 videos scraped, 12 proposals applied, design-generator skill created
View →
2026-04-28
Session Recap
DA-stream Lane A SSE + DA20 memory injection + DA21 action proposals + DA23 effectiveness leaderboard
View →
2026-04-27
Session Recap
DA13-DA18 AI panel skill substrate (picker, feedback, default, compose) + three-layer no-emoji enforcement
View →
2026-04-27
Session Recap
Sprint C R4+R5 chrome rename, DA12 slash menu, DA11.b voice pivot to local whisper.cpp
View →
2026-04-26
Session Recap
10/10 sweep — kernel errors silenced, trigger-health UTC bug, Karpathy on 5 skills, prospect-builder 7→10
View →
2026-04-26
Session Recap
Hybrid Detail AI complete — Lane A caching + Lane B SDK + tier allowlist + lane pill (DA5–DA10)
View →
2026-04-26
Session Recap
C-Sprint complete (C5–C8) · R2 shape mode · Detail AI primitive shipped
View →
2026-04-26
Session Recap
K-Sprint closed (K9, K10) · Tasks Kanban consolidated · C-Sprint launched (C1, C2)
View →
2026-04-22
Briefing
Fix Glide payment after 273 Baltic ends — two failures, MrNature OS at suspension risk
View →
2026-04-21
Briefing
Send Christopher Rodgers follow-up tonight — 8 days cold, 4th consecutive briefing
View →
2026-04-20
Briefing
Investigate and resolve the Glide quota breach tonight — RadQode apps offline
View →
2026-04-19
Briefing
Send the Christopher Rodgers follow-up before tonight — last desk window before field week
View →
2026-04-17
Briefing
Fix Glide payment ($301.90) before 9:30 AM field departure
View →
2026-04-16
Briefing
Send Christopher Rodgers follow-up before leaving for 725 Lafayette
View →
2026-04-15
Briefing
Fix Glide payment ($301.90 failed twice) before account suspends
View →
2026-04-14
Briefing
Follow up with Christopher Rodgers — discovery call done, 24-hour window closes today
View →
2026-04-13
Briefing
Prep the Christopher Rodgers discovery call — 2:30pm today, 2-hour window after field work
View →
2026-04-12
Briefing
Resolve Glide payment failure ($301.90) before the Hub goes dark — then prep the discovery call
View →
2026-04-11
Briefing
Prep the Christopher Rodgers discovery call before Monday — clear Saturday, full week ahead
View →
2026-04-10
Briefing
Christopher Rodgers discovery call at noon — first inbound pipeline call from radqode.com
View →
2026-04-09
Briefing
Prep the Christopher Rodgers discovery call — radqode.com live, booking confirmed
View →
RadOS
RadOS
Personal AI Operating System
The OS itself. Claude Code + Glide + Make.com infrastructure that runs all six ventures. Commands, agents, skills, memory, and harness all live here. Everything else is downstream of RadOS.
Claude CodeGlideMake.comGitHub
RadQode
RadQode
AI Operating Systems for Service Businesses
Productized RadOS builds for service business clients. Custom Astro + Supabase + Cloudflare stack. MrNature (landscaping) is Phase 1 free client. BackyardExpert (Alexander Rivera, NYC construction) is first paid build — $1K deposit pending.
AstroSupabaseCloudflareClient: Alexander
TaoWild
TaoWild
Outdoor Education & Nature Connection
Programs and content around outdoor education, Taoist nature philosophy, and wilderness skills. iOS app (Adaku, 97 Swift files — undocumented). 113 TaoWild Swift files pending commit.
iOS AppContentPrograms
RadicleBotany
RadicleBotany
Botanical Knowledge & Garden Consulting
Deep botanical expertise — taxonomy, ethnobotany, seasonal phenology, foraging safety. 108 uncommitted changes pending V2.0 tag. The Botanist skill provides cross-venture botanical intelligence.
Botanist Skill108 pending commitsV2.0
RadicleHerbal
RadicleHerbal
Herbal Products & Wellness
Herbal product development and wellness content. Leverages the Botanist skill for plant knowledge. On-demand operations — no active build sprint currently.
ProductsContentBotanist Skill
RadicleTerra
RadicleTerra
Land Stewardship & Ecosystem Work
Land care, ecosystem restoration, and environmental content. On-demand operations. TradePath (NYC contractor directory) is a potential adjacent venture being validated.
Land CareEcosystemTradePath adjacent
Commands
52 commands · .claude/commands/
Commands are slash commands invoked in the Claude Code terminal as /command-name. Each is a Markdown file in .claude/commands/ with a YAML frontmatter description. Commands are the primary interface for operating RadOS — they encode step-by-step instructions that Claude executes against the live system. 52 are RadOS custom commands (the local .md files, listed below). The Glide Commands table shows 102 rows total — the additional rows are Claude built-in commands (/help, /clear, etc.) and plugin commands (vercel:*, hookify:*, etc.) that auto-registered via the command-sync rule.
Operations
7 commands · Daily/Weekly cadence
/session-openStart a new work session — read the last session recap, show what's still open, set context for today
/session-closeEnd-of-session close — full recap of work done, save to _Sessions/, write Glide Report + Session row, surface next action
/briefingMorning briefing — yesterday's Claude work, today's Tasks from Glide, calendar, priority emails, venture momentum
/health-checkSystem health scan — check payments, domains, tasks, Glide quota, git status, and security alerts
/sub-checkMonthly subscription and payment audit — billing due dates, failed payments, burn rate, cancellation candidates
/task-reviewWeekly task review — apply DRIP Matrix, flag stalled tasks, output a clean priority stack for the week
/dispatch-checkProcess the Dispatch queue — execute queued commands and write results back
Intelligence
5 commands · Inbound signals, analysis
/email-monitorMonitor Gmail for actionable emails, create prioritized Tasks and Reports in Glide automatically
/email-reportRun an email response report for RadOS outreach campaigns and save to Glide Reports table
/venture-pulseVenture health check — single venture deep-dive or cross-venture snapshot
/live-contextFetch live environmental context for a venture — weather, season, daylight, calendar, and custom signals
/brainstormBrainstorm next automations, integrations, or workflows to build
Pipeline
6 commands · Lead-to-client lifecycle
/research-leadsResearch leads across all RadOS ventures and active projects — web search, deduplication, formatted output
/lead-ingestEnrich raw leads and write them to the Glide Leads table via API
/lead-qualifyScore leads based on engagement signals, promote qualified leads to Contacts, clean dead leads
/outreach-draftGenerate personalized outreach emails for leads in Glide and write email content back to the table
/follow-upScan for stale outreach leads and draft follow-up sequence emails
/campaign-launchLaunch a new marketing campaign — create brief, research leads, prep outreach pipeline
Content & AutoLearn
9 commands · Knowledge acquisition, content creation
/autolearnMaster orchestrator for the autonomous YouTube intelligence loop — monitor, scrape, analyze. Chains all steps.
/youtube-monitorPoll RSS feeds for new videos from watched channels and write to Glide YouTube table
/youtube-scrapeScrape a YouTube video's transcript and metadata, analyze content, write to Glide
/youtube-analyzeRun knowledge extraction on scraped YouTube videos — score relevance, generate improvement proposals
/youtube-applyReview and approve or reject Harness improvement proposals from all sources
/youtube-transcriptFormat raw YouTube transcripts into styled HTML and write to Glide
/youtube-searchSearch YouTube by keyword using yt-dlp — returns structured results for scraping or NotebookLM upload
/anthropic-monitorPoll Anthropic official sources for Claude updates — GitHub releases, status feed, blog, release notes
/content-ideasContent engine — batch ideas (no arg) or production brief (with topic arg)
Client / RadQode
9 commands · Prospect-to-client pipeline
/prospect-researchDeep automated research on a specific lead — scrape website, search reviews/portfolio/team, extract structured data
/prospect-previewGenerate a 5-section teaser visualizer for a prospect — branded HTML deployed to Cloudflare
/prospect-proposalGenerate a full proposal, schema spec, SaaS map, pricing, and 12-chapter OS Blueprint from intake data
/context-installInteractive onboarding installer for RadQode clients — populate OS context from questionnaire, docs, website, or paste
/intake-checkMonitor the Intake table for completed forms and dispatch proposal generation
/client-onboardOnboard a new RadQode client — create Drive workspace, generate SOW, create Glide record, draft outreach
/client-projectActive RadQode client project status — phase, deliverables, next milestone, blockers
/deploy-vizCommit and push visualizer changes to trigger Cloudflare Pages deploy
/backyard-expert-updateExecute queued BackyardExpert site changes from a Glide Dispatch payload — edit Astro components, deploy to Cloudflare
Harness
4 commands · System coherence & self-improvement
/khipuSystem coherence orchestrator — audit all RadOS layers against the eight Khipu architectural principles, score each, identify gaps, and plan or execute improvements
/command-auditAudit all commands against the 5-domain framework and Skills Trap anti-patterns
/command-evalEvaluate the output quality of the last command and log to the Harness
/friction-dashboardGenerate a friction trend dashboard — HTML report of recurring session failure patterns from .rados/friction-memory.json
Utilities
9 commands · Tools and setup
/invoice-createGenerate a new invoice from the RadOS template, create the PDF, and log to Glide Invoices table
/manualLiving Docs — add, update, audit, review, or deprecate content in Manual, Commands, Glossary, and other registered knowledge tables
/notebooklmConnect Claude Code to NotebookLM — create notebooks, upload sources, get analysis, generate deliverables at zero token cost
/infographicGenerate a NotebookLM infographic for any RadOS topic — full pipeline from source upload to PNG download
/visual-createBuild a dark-themed HTML reference visual and export as PNG — for learning, system understanding, and RadOS documentation
/playgroundCreate an interactive HTML playground for any topic
/route-dayOptimize today's MrNature job site driving order — geocode addresses, calculate optimal route, output Google Maps link
/gws-setupInstall and configure Google Workspace CLI for Claude Code integration
/glide-setupConfigure Glide tables and schema for RadOS
Retired
3 commands · Use replacements below
/content-briefUse /content-ideas [topic] insteadretired
/opportunity-scanUse /email-monitor for inbound signals and /follow-up for stale outboundretired
/snapshotUse /venture-pulse (no argument for cross-venture snapshot)retired
Agents
10 agents · .claude/agents/
Agents are specialized Claude subprocesses launched via the Agent tool. Each has a focused domain, a system prompt, and defined tools. They can run in the foreground (results inform next steps) or background (fire-and-forget). Pre-load at session start with claude --agent {name} for full context from token 1.
Ops
Daily Operations
Morning briefing, email monitoring, task management, session management, weekly reviews, subscription auditing, revenue opportunity scanning, and venture snapshots.
Content
Content & Intelligence
Orchestrates content creation workflows — batch ideation across ventures and YouTube transcript research. Chains /content-ideas and /youtube-scrape with venture strategy context flowing between steps.
Client
Client Relationships
Manages active client relationships, project delivery, proposals, scoping, and communication. MrNature, RadicleTerra jobs, RadQode projects. Primary agent for Alexander / BackyardExpert work.
Strategist
Strategic Intelligence
High-level strategic intelligence across all six ventures. Connects dots between opportunities, allocates attention and resources, identifies leverage points, and makes the hard calls about priority.
Design
Visual Intelligence
Sacred geometry architect. Creates production-grade visualizers, UI systems, and branded assets using mathematical design principles — Fibonacci proportions, golden ratio, color frequency science.
Sensei
Personal Mentor
Personal mentor and philosophical advisor. Guides clarity on life direction, personal alignment, discipline, health, and learning. Speaks with directness and depth — not a business optimizer, a wisdom layer.
System
Deployment & Infrastructure
Orchestrates deployment workflows and Glide schema creation. Multi-step processes that chain git operations, Cloudflare deploys, Glide API calls, and Hub updates in sequence.
Pipeline
Lead-to-Client Lifecycle
Manages the full lead-to-client lifecycle across all RadOS ventures — research, ingest, outreach, follow-up, qualification, conversion.
Templist
RadQode Client OS Templates
RadOS template architect for RadQode client delivery. Takes the RadOS system architecture and adapts, customizes, and deploys a client-specific operating system — schema, commands, agents, connections, automations.
Architect
System Coherence
RadOS system coherence evaluator. Audits all architectural layers (Agents, Skills, Commands, Harness, Connections, Automations, Glide tables) against the eight Khipu organizational principles.
Skills
6 file-backed · 14 total (8 Glide-only AI skills) · .claude/skills/
Skills are deep domain expertise modules loaded progressively via Claude's context system. Only the YAML frontmatter loads initially (preserving context budget). The skill body loads on trigger. Correct structure: skill.md = process only · references/ = brand voice, examples, rules. All 6 current skills are candidates for a references/ audit — monolith structure lowers activation rate to as low as 20%.
Content
Botanist
RadicleBotany · RadicleHerbal · TaoWild
Deep botanical knowledge — taxonomy, ethnobotany, seasonal phenology, foraging safety, and cross-venture plant intelligence. Shared by all three nature ventures.
⚠ audit: references/ subfolder
System
Archivist
RadOS · All Ventures
Notion sync, documentation quality, decision logging, memory persistence, and RadOS knowledge base structure. Institutional memory for the operating system.
⚠ audit: references/ subfolder
Operations
Financial Analyst
RadOS · Cross-venture
Financial health reporting, subscription auditing, burn rate analysis, and revenue tracking across all ventures. Powers /sub-check and /venture-pulse financial sections.
⚠ audit: references/ subfolder
Pipeline
Lead Researcher
RadQode · All Ventures
Deep prospect research framework for /prospect-research and /research-leads. Web scraping, profile extraction, ICP scoring, and structured lead output for the RadOS pipeline.
⚠ audit: references/ subfolder
RadQode
Prospect Builder
RadQode
Generate OS Blueprint proposals, branded preview visualizers, and schema specs for RadQode prospects. Bundles templates, design system, and sacred geometry layout patterns.
⚠ audit: references/ subfolder
Content
YouTube Learn
RadOS · AutoLearn pipeline
Autonomous YouTube video scraping, transcription, scholarly HTML generation, and Glide write pipeline for the AutoLearn knowledge system.
⚠ audit: references/ subfolder
AutoLearn System
_AutoLearn/ · Daily 5am
28Channels Watched
48Extractions
48Proposals Applied
7.25Avg Relevance (Last Run)
12New Videos (Last Run)
12Anthropic Updates (Last Run)
How AutoLearn Works
AutoLearn is a three-stage autonomous intelligence pipeline that runs daily at 5am via a scheduled trigger. It monitors 28 YouTube channels (via RSS), scrapes new video transcripts, extracts operational knowledge, generates improvement proposals for RadOS commands, and writes all results to Glide. Human review happens once via /youtube-apply.
Stage 1 — Monitor: Polls RSS feeds for 28 channels + 3 Anthropic GitHub feeds (Claude Code CLI, SDK Python, SDK TypeScript) + Anthropic blog. Writes new entries to the Glide YouTube/Intel table.
Stage 2 — Scrape: For each new video row, fetches the full transcript via youtube-transcript-api. Caches locally in _AutoLearn/extractions/. Skips already-cached videos. For GitHub releases, fetches full release notes via WebFetch.
Stage 3 — Analyze + Diff: Processes all scraped transcripts — scores relevance 1-10, extracts actionable knowledge, and generates typed proposals (command_modify, memory_entry, strategy_update). For command_modify proposals, reads the actual target file and writes a concrete diff. Proposals land in _Harness/adaptations/pending/ with Karpathy-style eval metrics for feedback loop tracking.
Last Run — 2026-04-04
Full pipeline: Monitor → Scrape → Analyze → Diffs → Report. Status: COMPLETE. Quota used: 42 mutations (~8,740 remaining this month). 10/12 analyzed videos scored 7+ relevance. 8 proposals generated — all approved and applied this session. Key sources: Mark Kashef (Claude Code hidden features, 9/10), Nick Puru (AutoResearch loop, 9/10), Simon Scrapes (Skills architecture x2), Jack Roberts (Claude roadmap: Kairos/Dream/UltraPlan), Nate Herk (500 workflows framework).
View Apply Report →
Harness — Self-Improvement
_Harness/ · _Harness/adaptations/
The Harness is RadOS's self-improvement layer. It captures three types of signals: friction events (where Claude failed), improvement proposals (from AutoLearn analysis), and command evals (quality scores on command outputs). Signals accumulate in Glide and .rados/friction-memory.json. Proposals land in _Harness/adaptations/pending/ and move to applied/ or rejected/ after /youtube-apply.
48Total Applied
0Pending
5Friction Patterns
×5Top Pattern Count
Active Friction Guardrails
×5Never launch agents, pipelines, or bulk operations before outputting a numbered plan and receiving explicit 'go' confirmation from the user.
×4Always fetch the live Glide table schema and confirm column key mappings before any write operation. Never assume keys from memory or prior sessions.
×3When generating HTML for Glide tables, always produce textbook-depth analytical content with 3-4 full paragraphs per section, minimum 5000 chars for video/intel rows.
×3Before using the Glide API, assess whether the operation could be done in the Data Editor. Reserve the API for bulk operations and automated workflows.
×2Never assume tool credentials need rotation without first checking: env var set, temporary network issue, key prefix match. Diagnose before escalating.
Tech Stack
Exact stack from CLAUDE.md
| Layer | Tool | Notes |
|---|
| Primary IDE | AntiGravity (Google) | Gemini 3.1 in Agent panel, Claude Code in integrated terminal |
| AI — Execution | Claude Code (claude-sonnet-4-6) | File edits, deploys, API calls, commands. Current model: claude-sonnet-4-6 |
| AI — Design/Planning | Gemini 3.1 Pro | Design decisions, content review, planning, analyzing existing code |
| Frontend Framework | Astro | output: server, Cloudflare adapter — for RadQode client builds |
| Styling | Tailwind CSS | Used in Astro builds |
| Deployment | Cloudflare Pages | Via wrangler CLI, npm run ship. Visualizers at rados.pages.dev |
| Database | Supabase | PostgreSQL + RLS, for RadQode client builds only |
| Hub / App | Glide | Internal OS dashboard, 21+ tables. App ID: 1CW7jf1VRWQgQi1lpgvg. 10K mutations/month. |
| Automation | Make.com | Webhooks, scheduled workflows, thin scheduler only |
| Email | Resend | Client notifications, contact forms. 3K free/month. |
| Domain / DNS | Cloudflare Registrar | At-cost domains ~$10-15/yr. All 7 venture domains migrated from GoDaddy 2026-03-18. |
| Version Control | GitHub — RadicleApps/RadOS | Main branch. gh auth not configured — blocks PR creation. |
| Google Workspace | GWS CLI (gws) | Gmail, Calendar, Drive, Sheets, Docs. Free reads — no API quota burn. |
| Knowledge Base | NotebookLM | RadOS Brain notebook. Used for AutoLearn infographics and semantic search. |
Dual-model workflow: Gemini (Agent panel) handles design decisions, content review, and planning. Claude Code (terminal) executes all file edits, runs deploys, makes API calls, writes code. Use @filename in the Agent panel to give Gemini file context. Take Gemini's suggestions to Claude Code to execute. Ship with npm run ship.
System Architecture
5 layers · Khipu Framework
External
External Services (5 bridges)
Gmail (MCP) · Make.com (scheduler) · Anthropic/Claude API (Glide AI) · Glide API (queryTables/mutateTables) · Cloudflare (Pages, DNS, email routing)
Commands
Slash Commands (49 active)
Operations · Intelligence · Pipeline · Content/AutoLearn · Client/RadQode · Harness · Utilities — the verbs of the system. Each is a Markdown instruction file.
Agents / Skills
Agents (10) + Skills (6)
Specialized subprocesses with domain expertise. Agents orchestrate multi-step tasks. Skills provide deep knowledge via progressive disclosure. Pre-load with claude --agent {name}.
Data
Glide Tables (21+) + Local Files
Glide: Tasks, Leads, Contacts, Pipeline, Reports, Sessions, Videos, Harness, Dispatch, Intake, Invoices, Services, Commands, Agents, Skills, Manual + more. Local: _Sessions/, _Briefings/, _AutoLearn/, _Harness/, _Cache/, _Reports/, memory/.
Memory
Persistence Layer — memory/ + _Sessions/ + Harness
memory/MEMORY.md (40+ entries: user, feedback, project, reference) · _Sessions/ (session recap archive) · .rados/friction-memory.json (pattern tracking) · _AutoLearn/extractions/ (48 knowledge files). Kairos (when it ships) will read this layer continuously.
Khipu Framework: The 8-principle architectural system derived from Inca information architecture research (2026-04-02). Powers
/khipu command and Architect Agent. Evaluates system coherence across all layers. Key current gaps: Snapshot table (pre-aggregated stats, P2), VENTURE-CONTEXT.md files (P3). Full framework at
_Builds/KHIPU-RADOS-ARCHITECTURE.md.
Memory System
memory/ in ~/.claude/projects/
Four Memory Types
User — who Chris is, expertise, preferences. Shapes how Claude tailors explanations and recommendations.
Feedback — corrections and confirmations from prior sessions. The most important type — prevents repeating mistakes and confirms working approaches.
Project — ongoing work, decisions, goals, blockers. Decays fast — always includes why and how to apply.
Reference — where things live in external systems (Glide table IDs, Notion page IDs, API configs, credentials).
Current Memory Index
18 project entries (BackyardExpert, Scholarly HTML system, Strategic Review, Glide schema, Agentic architecture, Marketing engine, RadQode builds, Khipu framework, Skills architecture, Claude roadmap signals + more)
8 reference entries (Gmail labels, Reports table API, Notion hub, GitHub repo, Cloudflare account, iMessage channel, Credentials table, Claude Code roadmap)
20 feedback entries (core behavioral constraints that persist across sessions). Full index at memory/MEMORY.md.
Keeping this Hub current: This file is generated from live RadOS folder state. To regenerate with updated counts, session data, and Harness stats, run: python3 _Scripts/generate-hub.py (to be built) or ask Claude to rebuild the Hub during any session. The Hub is not live-updating — it reflects the state at generation time (2026-04-04). Stats like command count, agent count, and session info are read from the actual .claude/ directory and _Sessions/ folder.