Ops Pulse
All clear. No unread notifications.
Pending Approvals
No pending approvals
Automations will stage drafts here for your review before sending.
ForgePulse Live Feed
All clear. No pending notifications.
All clear. No unread notifications.
Automations will stage drafts here for your review before sending.
All clear. No pending notifications.
GOAL DUE: Mainframe Phase 1: Foundation deployed to mainframe.apexfoundry.co
4d left
ForgePulse Cron Trigger
No runs in 24h
ForgePulse Morning Ops Report
No runs in 24h
ForgeTracker Daily Backup
No runs in 24h
ForgeTracker MCP (Supabase Edge Function)
No runs in 24h
Netlify Intake: Custom Plugin
No runs in 24h
Netlify Intake: New Website + Hosting
No runs in 24h
Netlify Intake: New Website Only
No runs in 24h
Netlify Intake: New Website Only
No runs in 24h
Netlify Intake: WordPress Migration
No runs in 24h
QuickBooks Online MCP Integration
No runs in 24h
Retell AI Phone Receptionist
No runs in 24h
SWB MCP (Supabase Edge Function)
No runs in 24h
Vercel Auto-Deploy (Next.js)
No runs in 24h
Apr 6 6:00
Lesson: Building without deploying: six weeks of framework, zero operational capability live
Apr 6 5:00
Lesson: Dark mode must mirror light mode warmth, not invert to navy/void
Apr 6 4:00
Lesson: Mandatory clarification questions before any ForgeTracker/ForgeMemory write operation
Lesson: Brand v7.0 requires propagation across 8 skill files and 12 active builds with UI
Apr 6 3:00
Lesson: Brand identity requires iterative visual exploration with the operator, not spec-first design
Lesson: $100M visual standard is the permanent manufacturing specification for all ApexFoundry surfaces
Apr 6 2:00
Lesson: mcp-forge production line targets Node.js npm but all deployed MCPs are Supabase Edge Functions -- full architecture mismatch
Apr 6 1:00
Lesson: ForgeTracker/ForgeMemory are passive recorders not active enforcers -- same process failures repeat every session
Lesson: Four-tool governance framework: Claude AI / Claude Code / Claude in Chrome / Cowork have strict non-overlapping roles
Apr 6 0:00
Lesson: 10 consecutive clean Mainframe builds -- one-phase-at-a-time feedback loop is the system
Lesson: Figma status was wrong three times in one session -- clarify before changing operational state
Lesson: Memory edit incorrectly listed Figma as dropped -- it is active and used for all design work
Apr 5 23:00
Lesson: Plugin/skill audit: 37% of Claude Code plugins serve dead projects, critical Vercel upgrade and react-best-practices missing, monthly toolchain audit needed
Lesson: BATCH-PROMPTS.md is not a pre-written queue -- prompts are delivered one at a time with feedback loops between each
Apr 5 22:00
Lesson: Self-upgrade protocol: Claude operating configuration version-controlled in ForgeMemory with 10 trigger conditions
Lesson: Claude in Chrome cannot access local filesystem -- all code must be inline in prompts, never file path references
Lesson: Edge Function deliverables must go through the full forge feedback loop -- artifact, automation, build status, memory
Lesson: Always give Claude in Chrome prompts for browser work -- never CLI commands or manual instructions
Lesson: ForgeTracker builds table is actually build_status -- spec says builds but Supabase table is build_status
Apr 5 21:00
Lesson: 33 Chrome extensions exist but none were built through ext-forge -- all are prototypes, not enterprise products
Apr 5 20:00
Lesson: Correction overcorrected: Phase 4 was actually built but got downgraded without verifying on disk
Lesson: Never give vague browser instructions when Claude in Chrome prompts are the standard -- always provide the full prompt with URLs and build context
Lesson: Combine related dashboard tasks into single Claude in Chrome prompts -- reduces operator context switching
Lesson: Claude in Chrome prompts must include full navigation URL as first instruction -- never assume user is already on the page
Lesson: Two interfaces to one backend -- Mainframe and Webvision Platform share all infrastructure, eliminating divergence
Apr 5 19:00
Lesson: A/B Testing Requires Server-Side Implementation and Statistical Rigor
Lesson: Build IP Not Reseller Margin -- White-Label is a Ceiling Not a Floor
Lesson: Platform Lock-in Through Daily-Use Tools
Lesson: Phase 2D Pending Approvals: seventh consecutive clean build -- OpenClaw gap analysis directly produced a shipped phase
Lesson: Phase 4 logged as complete but was never built -- prompt written is not phase complete
Lesson: Forge Context Ingestion Protocol validated: 3 runs, 56 tool calls, 46+ items, zero conflicts, intelligence fed directly to active build
Apr 5 18:00
Lesson: Always run formal gap analysis when replacing a deprecated system -- spec vision misses predecessor capabilities
Lesson: OpenClaw planning intelligence survived platform deprecation -- operational thinking transfers, platform config is disposable
Lesson: Competitive analysis confirms Mainframe undercuts all 7 commercial platforms while delivering broader feature set -- build-not-adopt is a verified cost advantage at 2-operator scale
Lesson: Forge Context Ingestion Protocol bridges the gap between conversation intelligence and operational state -- without it, research conversations are disposable
Lesson: RLS disabled was a known state that no session escalated to production blocker -- security inspection must happen before deploy, not after
Lesson: Deep Research as pre-build system audit tool -- structured commission with acceptance criteria produces dramatically better output than ad-hoc questions
Apr 5 17:00
Lesson: Session handoff pattern validated: full prompt delivery worked first try after previous session fixed seven handoff failures
Lesson: Phase 4 Financials: seventh consecutive clean build -- most complex phase yet with 9 deliverables, 6 components, 4 API routes
Lesson: Handoff gave Claude Code a file reference instead of the full prompt -- BATCH-PROMPTS.md is storage, not delivery
Apr 5 16:00
Lesson: Phase 3B: wizard transformed into guided consultation -- 12 types, discovery questions, service guide -- sixth consecutive clean build
Apr 5 13:00
Lesson: Phase 3 (Clients tab + wizard + 10 onboarding templates + 6 API routes) built clean -- fifth consecutive
Apr 5 12:00
Lesson: Session handoff keeps failing -- ForgeTracker stores history but not executable next-action context
Lesson: React stale closure in context providers: always use functional setState in callbacks consumed through context
Lesson: Phase 2C: fourth consecutive clean build -- soft delete, light/dark themes, operator scoping, file attachments, brand onboarding all landed
Apr 5 11:00
Lesson: pulse_add_custom_watcher MCP tool missing entity_scope parameter -- causes wrong Slack routing on all custom watchers
Lesson: Claude Code handoff prep checklist: CLAUDE.md update, Supabase prep, and BATCH-PROMPTS.md verification must all happen BEFORE telling operator to build
Lesson: Phase 2B (messaging + visual + onboarding) built clean -- third consecutive clean phase for Mainframe
Apr 5 10:00
Lesson: The forge system is a digital product factory -- same production lines that build software produce sellable assets at zero marginal cost
Lesson: Every operational workflow must be audited for dual-business entity awareness -- data separation is not enough, logic must be entity-specific
No active recurring operations.