Operations

Building without deploying: six weeks of framework, zero operational capability live

Over six weeks, we built a complete Mainframe UI (6 tabs, 90 files, 40 API routes), designed 40 database tables, coded 145 MCP tools, wrote a 1759-lin...

Dark mode must mirror light mode warmth, not invert to navy/void

When implementing dark mode for a light-first brand, do NOT invert to a traditional dark palette (deep navy, void black). Instead, take the light mode...

Mandatory clarification questions before any ForgeTracker/ForgeMemory write operation

Before writing ANY data to ForgeTracker or ForgeMemory, Claude must present clickable clarification questions using ask_user_input_v0. This applies to...

Brand v7.0 requires propagation across 8 skill files and 12 active builds with UI

Brand tokens v7.0 requires propagation across 8 Claude.ai skill files that have hardcoded v6.0 values (#00C8EE, void #010102, Geist). The skills are: ...

Cross-Platform AI Adoption Validation: OpenAI vs Anthropic Usage Gaps

Cross-platform validation: OpenAI published work usage report (January 2026) showing same adoption gaps as Anthropic Economic Index. ChatGPT at 900M+ ...

Anthropic US Use Case Distribution and Country Usage Index

Anthropic country-level data: US ranked 5th/116 countries with 4.58x usage index (222K observations). Singapore leads at 5.53x, Israel 5.20x, Luxembou...

Brand identity requires iterative visual exploration with the operator, not spec-first design

Five rounds of brand concept iteration were needed to land on the correct visual direction. The process: (1) three dark-void concepts rejected -- all-...

$100M visual standard is the permanent manufacturing specification for all ApexFoundry surfaces

Every ApexFoundry output must look like a $100M company to command $100K+ engagement checks from enterprise buyers. This is not aspirational -- it is ...

mcp-forge production line targets Node.js npm but all deployed MCPs are Supabase Edge Functions -- full architecture mismatch

Full audit of C:\apexfoundry-forge revealed that the mcp-forge production line targets the wrong architecture for every MCP actually deployed in this ...

Anthropic AI Adoption Rate by Occupation -- Market Opportunity Index

Anthropic published AI usage data covering 974 occupations across THREE dimensions: (1) Task Coverage -- how broadly AI covers tasks, (2) Automation -...

ForgeTracker/ForgeMemory are passive recorders not active enforcers -- same process failures repeat every session

The ForgeTracker and ForgeMemory system has three structural failures that cause the same process errors to repeat every session: (1) MEMORY CONSOLIDA...

Four-tool governance framework: Claude AI / Claude Code / Claude in Chrome / Cowork have strict non-overlapping roles

Each tool in the ApexFoundry stack owns a distinct function. Claude AI is the architect: writes specs, prompts, QC analyses, session management, intel...

CURRENT STATE: Phases 1-6 UI COMPLETE. Brand v7.0 COMPLETE. MCP Edge Functions DEPLOYED. Vercel deploy IN PROGRESS. LIVE RIGHT NOW: - ForgeTracker v3.2: 127 tools (102 CRUD + 25 agent runtime) at uph...

Daily Operations Playbook v1.0. The operating system for ApexFoundry + Webvision. Five phases, every session. TRIGGERS: - "Give me the pulse" -> Phase 1 - "Run the revenue triggers" -> Phase 2 - "Run...

ApexFoundry dark mode is WARM CHARCOAL, not navy/void. Principle: "light mode, just dimmer -- same room, lights turned down." Surfaces: #222226 (bg-primary), #2A2A2E (bg-surface), #323236 (bg-elevated...

The market intelligence system has THREE data layers, all cross-validated across platforms. SUPPLY SIDE: occupation adoption rates (974 occupations x 3 dimensions from Anthropic Economic Index: task_c...

Market Intelligence is a ForgeTracker subsystem (table 40) storing AI adoption data by occupation/industry as a queryable opportunity index. Source: Anthropic usage data (974 occupations, 3 dimensions...

MANDATORY PRE-BUILD GATE -- must complete before any code, TypeScript, Edge Function, or deliverable is produced. No exceptions. Skipping any step is a process defect. STEP 1: Identify production lin...

FOUR-TOOL ROUTING RULES (established 2026-04-06): CLAUDE AI: Strategy, specs, WORK-ORDER.md, CLAUDE.md, QC analysis, session management, ForgeTracker/ForgeMemory ops, intelligence scans, Edge Functio...

WordPress Studio is the local development environment for all WordPress plugin builds and client site testing. Charlie has it installed (v1.7.5). Integration with forge workflow: Claude Code builds pl...

DAILY recurring operation: Anthropic product intelligence scan + SYSTEM ANALYSIS. Not just a report -- when new Anthropic features are detected, Opus 4.6 must ANALYZE the full ApexFoundry system again...

Strategic direction: USE ALL OUR OWN STUFF. What we build replaces third-party tools and does better. Third-party tools are bridges during transition, not permanent dependencies. DROPPED (replaced by ...

Daily Anthropic scan April 5, 2026. CRITICAL FINDINGS:\n\n1. CLAUDE MYTHOS MODEL LEAKED: New model ABOVE Opus 4.6. Fourth tier called 'Capybara'. Dramatically better at programming and reasoning. CMS ...

Comprehensive audit completed April 5, 2026, corrected per operator. CURRENT: 55 Claude.ai skills (38 user, 7 public, 10 example), 19 Claude Code plugins, 5 agents, 4 hooks, 3 marketplaces, 18 MCP con...

Claude Code prompts are delivered ONE PHASE AT A TIME, not batched in a file. The workflow is: (1) Claude.ai writes one prompt for the current phase and presents it inline in chat. (2) Charlie pastes ...

SELF-UPGRADE PROTOCOL for Claude.ai system instruction and memory edits. The operating configuration for Claude evolves as the business evolves. Two artifacts are version-controlled: (1) System instru...

MEMORY EDITS v1.0 (2026-04-05). 30/30 slots used. Boot loader entries (lean pointers, not the brain): #1 no emojis, #2 partner not tester, #3 QC background, #4 ENTJ forge identity, #5 match intensity,...

SYSTEM INSTRUCTION v1.0 (2026-04-05). Location on disk: C:\apexfoundry-forge\CLAUDE-AI-SYSTEM-INSTRUCTION.md. Installed in Claude.ai Settings > Profile > Custom instructions. Contains: identity (Charl...

The memory system is the turbocharger for every session. Three layers work together: (1) memory_user_edits (30 slots) is the boot loader -- lean pointers that tell Claude who Charlie is, how to behave...

Claude in Chrome CANNOT access Charlie's local filesystem. It runs in the browser context only. When writing Claude in Chrome prompts, NEVER reference local file paths like C:\apexfoundry-forge\... or...

Complete build inventory as of April 5, 2026 (nine phases complete, all clean): LAYOUT: TopBar, Sidebar (5 tabs with live badge counts), MobileNav, Shell, ConnectionIndicator. OPS PULSE TAB: PendingAp...

Five protocols handle all system communication. NO external automation platform needed (no n8n, Zapier, Make). (1) MCP: how AI operators interact with data. Three layers: ForgeTracker (99 tools, raw C...

Three deployment paths from forge to HostArmada VPS, all proprietary, no external CI/CD needed: PATH 1: WP-CLI PROXY (plugin deployment to client sites) Flow: Claude Code builds plugin -> QC gate pas...

ApexFoundry operates as an ENTJ company. Systems thinking, production lines, and leverage in everything. Manufacturing mindset: nothing ships half-built, quality and safety standards are non-negotiabl...

Every Chrome extension in the Apex Extension Suite must be enterprise grade, pushing the boundaries of what is possible for MV3 as of April 5, 2026. ApexFoundry sets the standard. Requirements: (1) Bu...

NEVER mark a phase complete until Charlie confirms the Claude Code build ran clean. A prompt existing in BATCH-PROMPTS.md means 'ready to build,' NOT 'built.' The feedback loop (forge_update_build_sta...

WordPress maintenance in Mainframe uses a two-layer architecture that eliminates SSH entirely. Layer A: WHM/cPanel REST API integration via forge-whm-api Edge Function for monitoring (account resource...

10 gaps between OpenClaw capabilities and Mainframe replacement, identified April 5 2026. Resolution plan with LLM bucket assignments: GAP 1 (HIGH): Self-improvement loop. Build forge-learning-loop E...

ForgeSwarm is a custom Postgres task bus built on ForgeTracker tables + Edge Functions + direct LLM API calls. No external framework dependency. Tasks table is the orchestration queue (status: queued-...

Verified April 5 2026: Mainframe production infrastructure costs $52-95/month. Supabase Pro $25/mo (eliminates auto-pause, daily backups, 2M Edge Function invocations). Vercel Pro $20/mo (1TB bandwidt...

CRITICAL: RLS is disabled on all 39 tables with anon key exposed in browser JavaScript. Any client-side code inspection reveals the key, enabling full CRUD on all tables via PostgREST API. RECOMMENDED...

For large builds with many components, work is split into batches of 3-5 items. Before ANY batch or phase: Claude.ai completes the pre-build prep checklist (update CLAUDE.md, verify BATCH-PROMPTS.md, ...

MANDATORY PRE-BUILD PREP CHECKLIST (Claude.ai must complete ALL steps before giving the operator the Claude Code prompt): (1) READ CLAUDE.md from disk. Verify it matches reality. Update: mark complete...

The Claude Code prompt delivery pattern has two phases: PREP and DELIVER. PREP (Claude.ai does this): (1) Read CLAUDE.md from disk. Verify it matches reality. Update if needed: mark completed phases, ...

MANDATORY SESSION END CHECKLIST (do ALL before forge_end_session): (1) Update CLAUDE.md on Charlie's disk -- mark completed phases, update current prompt reference, add any new critical rules learned....

Pending items beyond Phase 4 build as of April 5, 2026: (1) ForgePulse daily-revenue-product watcher fires to wrong Slack channel -- needs entity_scope=apexfoundry. The pulse_add_custom_watcher MCP to...

Phase 4 is the Financials tab. Scope per SPEC.md Section 7.3 (line 368): (1) P&L Summary with month selector, revenue/expense totals, net income, MoM trends, entity breakdown, CSV export. (2) Per-Clie...

The Niche-to-Print concept reveals a reusable ApexFoundry pattern: GENERATE -> VALIDATE -> COMPILE. Instead of guessing what will sell, procedurally generate variations, deploy them as lightweight web...

Every Supabase schema change required for a Claude Code build phase must be delivered as a Claude in Chrome prompt. Never tell Charlie to "go run SQL in Supabase" or "go to the dashboard and do X." Th...

Three-stage deletion pattern for all records in Mainframe: (1) SOFT DELETE -- user clicks delete, item visually disappears, 10-second undo toast appears at bottom of screen. If undo clicked, item rest...

Operator permissions in Mainframe v1: Charlie (MAINFRAME_OPERATOR=charlie) sees ALL data across both businesses, business toggle shows ApexFoundry/Webvision/Both. Steven (MAINFRAME_OPERATOR=steven) se...

Every morning starts with a quick-win revenue production session. Charlie produces at least one sellable digital product per day using the forge system and Claude as build partner. Product types: prom...

ApexFoundry is the builder entity. It does everything Webvision does plus more. ApexFoundry builds software, tools, digital products, Chrome extensions, SaaS platforms, AI automations, consulting serv...

Whenever a task requires work in the Supabase Dashboard (running SQL, checking settings, enabling Replication, adding secrets, viewing API keys, creating tables, modifying columns, deploying Edge Func...

In Next.js 16.2.2, proxy.ts MUST be at the project root or at src/proxy.ts (same level as app/). It must NOT be inside src/app/. The exported function must be named proxy(), not middleware(). middlewa...

ApexFoundry Mainframe is the unified operational platform for both businesses, replacing Slack, QuickBooks, and OpenClaw with a single system. Three operators use it: Charlie (founder), Steven (partne...

Mainframe is the complete operating system for ApexFoundry and Webvision Group. It replaces Slack (alerts route to Ops Pulse natively, operator messaging within Mainframe), replaces QuickBooks (invoic...

Steven fully supports expanding beyond WordPress websites. As of April 5, 2026, both partners are aligned on: (1) Mainframe as the unified operational platform replacing Slack and QuickBooks, (2) expa...

Mainframe Phase 1 is COMPLETE. The app-forge workflow executed end-to-end successfully. Files on disk: SPEC.md (1759 lines, 48 sections), WORK-ORDER.md (11 sections, app-forge template), CLAUDE.md (ve...

ApexFoundry Mainframe Phase 1 COMPLETE and verified live with real ForgeTracker data. Next.js 16.2.2 on localhost:3000 (Vercel deploy pending). 22 files built: proxy.ts auth gate (deferred to pre-depl...

v0.2.0 SHIPPED. All 4 forge phases complete. Production ZIP: 361KB, 246 entries. 22 blocks total (7 v0.1.0 + 15 v0.2.0 including 2 child blocks for Tabs and Accordion InnerBlocks). 22 template pattern...

CRITICAL RULE: Claude.ai NEVER writes application code using computer tools (bash_tool, create_file, str_replace). Claude.ai's role is SETUP: write WORK-ORDER.md, update CLAUDE.md, prepare Claude Code...