Changelog.

2026-04-18 (Site polish — pricing, onboarding UX, agents page)

/framework page

• Removed redundant "3 ways in" heading so the cards lead the section
• Pricing tier alignment fixed — descriptions flex-grow and price row is fixed-height, so Get Starter / Coming soon / Let's talk buttons line up across all three cards
• Replaced testimonials with Founding Community section — real GitHub avatars via github.com/{handle}.png plus dashed + slots for future builders
• Removed post-purchase timeline section (low-signal noise)
• Integration logos switched from white silhouettes (brightness-0 invert) to colored logos on white rounded tiles — real brand colors, more recognizable
• FAQ refresh: dropped obsolete "upgrade Starter → Pro" question, added refund policy entry, clarified Pro launch status
• Starter CTA rerouted from /auth directly to /build so users complete pain/outcome/scope capture before checkout

/build starter onboarding

• Reload always starts fresh: localStorage draft cleared on mount (was resuming from the last step)
• Step 7 rebuilt from scratch — killed the SVG network mesh, floating tooltip card, dark terminal cosplay, and the "agents aligning…" gate that delayed Continue
• Replaced with a Configuration summary card + a Personalized CLAUDE.md preview rendered from the user's answers, plus a clean 7-agent row with avatars + roles
• ~200 lines of animation timers, event arrays, and network-node coordinates deleted
• Continue button no longer gated by completion of an animation

/memory-doctor email gate

• Email-capture card now overlays a blurred preview of the actual report (header + donut score + stats + top 3 insights) — the user can see the report exists behind the form
• Mobile/iPad polish: responsive card widths (max-w-md sm:max-w-lg), tighter stat cells on phones, autoComplete="email" + inputMode="email" for mobile keyboards
• Preview is aria-hidden + pointer-events-none so it doesn't steal focus

/agents page

• Full rewrite. Cards are now clickable and link to /resources/agents/{slug} docs pages
• Emoji swapped for CreatureAvatar inside a tinted halo in each agent's own color (cortex purple, loom violet, radar blue, etc.)
• Cortex gets featured row (full-width on md+) as Chief of Staff
• Killed the [Diagram placeholder] box. Replaced with a 4-step orchestration grid: Brain dump → Cortex routes → Specialist executes → You approve
• Added closing CTA card linking to /framework#pricing and /build

2026-04-17 (Messaging rewrite + content push)

Messaging

• Site-wide messaging refresh across landing and framework pages
• Security page shipped
• Image style guide documented for Gemini image generation (bold, vibrant, saturated — referenced before every image generation)

Content

• 3 new blog articles drafted and published
• Personal info sweep across articles: family names, API keys, internal identifiers scrubbed
• New feedback memory: scan blog for personal info before publishing

pSEO Retool

• 18-task implementation plan approved for Agent0 pSEO hub + 15 sub-pages + Scribe agent
• Spec + plan committed to artifacts/plans/
• Backlog populated with the 18 tasks, pending start

2026-04-12 (Session 2, marathon BIPS + drafts)

Build-in-Public Sync (BIPS) — End-to-End

• Phases 0-7 shipped in one session: schema + redaction, webhook pipeline, /pulse + /brain + /radar + /changelog public pages, approval-task flush, 30-day backfill
• Deployed 2 Supabase migrations to prod, 851 rows backfilled from 30 days of vault commits
• 50+ tests passing (redact 24, classify 11, HMAC 5, flush 10)
• No-LLM fallback path so system works with or without ANTHROPIC_API_KEY

Marketing Site

• /vault landing page shipped (Vault tier $129 pitch)
• /changelog now mirrors vault CHANGELOG.md with branded markdown rendering
• iPad + landscape responsive pass across nav, docs layout, card grids, chat bubble, alert banner
• Fixed 3 drift bugs in /resources (article-ideas missing from CLAUDE.md, missing top-level folders, hardcoded skill counts → drift-proof JSON configs)
• Fixed nested <button> hydration errors on /blog and /build

Vault Architecture

• SETUP.md written for Vault tier buyers (under-15-min path from repo invite to first /daily session)
• Postiz MCP server built + tested + moved to app repo (JS doesn't belong in vault)
• .mcp.json at vault root points to app-repo postiz by absolute path
• CLAUDE.md article-ideas/ added to canonical artifacts list + alphabetized
• CLAUDE.md extended with Build-in-Public Frontmatter section (private, public, publish_to, public_name, generate_image)

Design Docs + Plans

• artifacts/briefs/agent0-tenancy-architecture.md — 10-scenario threat model + schema sketch for hosted Pro multi-tenancy
• artifacts/plans/radar-pipeline.md — 7-stage Radar lifecycle design (captured → shipped)
• artifacts/plans/learning-layer.md — Shadow → Canary → Promote for skill evolution
• governance/Agent Wakeup Policy.md — enable/disable + conditional triggers + vacation mode for Cortex scheduled

Decisions

• Article routing: no cross-posting (one venue per article); image gen only when explicitly flagged
• BIPS trigger: GitHub webhook → Next.js API route (not GH Action)
• Tenancy: Supabase virtual FS + single managed ANTHROPIC_API_KEY, no BYO

Governance

• tasks/waiting-for-review/ holds 6 items for Maciek review
• 9 tasks archived this session

2026-04-11 (Session 4-5, marathon)

Hub Files Complete

• 16 hub files written with 14-section template (8 vault + 5 system + Radar + People + Runs)
• Each Codex-reviewed, findings fixed, governance aligned

Architecture Overhaul

• 7-agent model: removed "heartbeat" concept. Cortex runs interactive + scheduled.
• Schedule redesigned: Radar Scout (outbound) + Cortex (inbound), 3 runs/day
• Skills restructured: flat {category}-{name}/ format, 83 symlinks deleted
• Inbox killed: daily note is universal entry point, Quick Captures checkbox model
• Demand Radar pipeline: wired into /capture, governance, schedule, /sync
• Coordinator-heartbeat.md: full rewrite (5 roles, flat skill paths, memory writes allowed)

Governance Updates

• Cortex: can update ACTIVE.md, create projects, write decisions, update Task Board Backlog, write memory (all approval-gated)
• Radar: owns ideas/ pipeline, scheduled Scout permissions for radar/ and knowledge-base/repos/
• Hippocampus: mandatory challenge gate for idea activation, can validate radar/ signals
• Axon: added to governance, can review run traces
• Sentinel: role corrected to operations analyst

CLAUDE.md Audit

• 12 Codex findings fixed across all sections
• Vault structure updated (radar/, people/, flat skills)
• Protected files list corrected
• Context rules updated (dead targets removed, radar routing added)

Consistency Cleanup

• Pulse→Cortex scheduled across 20+ files
• inbox/→daily note Quick Captures across skills
• Memory files updated to match new architecture
• Codex final check: core system files clean

2026-04-06 (Session 3)

Hub File Template

• Designed 14-section hub file template: What This Is, How It Works, Folder Structure, Inputs, Outputs, Hard Bans, Definition of Done, Escalation, Skills, Agents, Health Signals, File Template, Connections, Live View
• Applied to Daily Journal and Ideas hub files

Kill Inbox

• Decision: inbox/ folder deprecated. Daily note is now the universal entry point.
• Implemented across 15+ files: CLAUDE.md, /daily, /capture, /tldr, governance, heartbeat, templates
• Physical inbox/ folder removed. assets/ shared directory created.
• Capture state model: - [ ] pending, - [x] processed in Quick Captures section

Ideas Pipeline Ownership

• Radar declared owner of ideas/ section (scoring, ranking, deduplication, activation recommendations)
• Hippocampus is mandatory challenge gate before activation
• Status lifecycle: captured, scored, challenged, activated, shipped, parked, archived
• Scoring enforced: values 1-3 only. Template defaults to status: captured
• Agent Permissions updated: Radar writes to ideas/, Hippocampus writes challenge memos, Cortex updates ACTIVE.md after approval

Product-Readiness

• "Maciek" replaced with "the human" in /daily, /capture, /tldr skills
• "Maciek's Entry" renamed to "Human Entry" across all templates and skills

Codex Integration

• Installed codex-plugin-cc (OpenAI Codex as adversarial reviewer for Claude Code)
• 5 adversarial reviews completed across Daily Journal and Ideas hub files

2026-04-03 (Session 2)

Repo Follow Pipeline

• Captured 10 GitHub repos to knowledge-base/repos/ with standardized 9-section digest format (frontmatter, what, features, architecture, why, suggested tasks, connections, README callout, recent activity)
• Created 6 new people files for repo authors
• Repo digests include embedded README and recent commits for visual browsing in Obsidian

Skills

• Added customer-research skill (v1.0.0) — pulled from upstream marketingskills repo (Corey Haines). JTBD, VOC, digital watering holes, persona generation. 77 total skills.
• Enhanced brainstorming skill with structured interview phase — 6 question categories (Purpose, Users, Constraints, Success Criteria, Edge Cases, Integration Points) with "interview complete" gate before proposals

Hooks

• Added memory-nudge.sh PostToolUse hook — session-aware memory consolidation nudge at 20/40 tool uses, suppresses if memory recently written

Memory

• Added feedback_always_update_kanban.md — must update both task file AND Task Board.md
• Added feedback_repo_digest_format.md — 9 required sections for repo digests

CLAUDE.md

• Updated skill count: 76 → 77
• Added customer-research to marketing skills table

2026-04-02

/capture Skill — Video Digest Pipeline

• Upgraded /capture to auto-detect video URLs (YouTube + TikTok) and trigger enhanced pipeline
• Added hype filter (3.4/5 calibration) — compares video suggestions against full vault system (skills, SOPs, workflows, agents, memory)
• Added verdict system: SKIP (already doing it / hype) → EVALUATE (worth reviewing) → IMPLEMENT (clear gap)
• Added people tracking — every content creator gets a people/ file, updated on each capture
• Added video-digest template for knowledge-base/ output
• Video URLs now route differently from standard URL captures

Apify MCP Server Integration

• Connected Apify MCP with TikTok scraper (clockworks/tiktok-scraper) and YouTube scraper (bernardo/youtube-scraper)
• Provides structured JSON: transcripts/subtitles, creator profiles, engagement metrics, comments
• Replaces defuddle for video content — defuddle can't extract JS-rendered transcripts
• Auth required on next session restart

New Feedback Memory

• "Always find structured, LLM-readable data sources" — don't default to defuddle/WebFetch for dynamic content

2026-04-01 (Session 2)

Scheduled Agents — Named Identity

• Renamed: Vault Heartbeat → Pulse, Nightly Entry → Chronicle, Weekly Reflection → Mirror, Monthly Reflection → Compass
• All 4 trigger prompts updated with new names in branches (agent/pulse-*), commits, and PR titles
• /sync skill updated to recognize both old and new naming patterns
• CLAUDE.md updated with named agents

Task Template Enforcement

• Heartbeat/Pulse trigger now enforces full task template (Feedback, Progress Log, Skills to Use sections)
• Fixed 4 existing task files to match template

Self-Improving Agent Spec

• Wrote conversation-driven feedback loop spec (artifacts/plans/self-improving-agent-spec.md)
• 4 layers: Detection → Storage → Application → Verification
• instances counter for repeat failures, escalation at 3+

Demand Radar Pipeline

• Created radar/ vault folder with Radar.md home and dataview query
• Created _templates/Signal.md for Obsidian Templater
• Scout agent designed but blocked by 4-trigger plan limit

Agent0 Website Master Plan

• Created artifacts/plans/agent0-website-masterplan.md
• Value prop shifted: "Build Your Brain" → "Ship With Agents"
• Value chain: Structure → Instruct → Orchestrate → Ship

2026-03-31 → 2026-04-01

Knowledge Base Audit + Skill Expansion (76 skills)

• Audited 5 GitHub repos (Everything CC, Awesome CC, GSD, UI/UX Pro Max, LightRAG)
• Created 4 new skills: strategic-compact, context-presets, ui-design-intelligence, ui-implementation
• Upgraded 3 CRO skills with visual/accessibility checklists
• Installed agnix (skill linter) + claude-rules-doctor (dead rule detector)
• Rewrote CLAUDE.md with full 76-skill catalog in categorized tables
• Archived 4 processed KB docs, cleaned up knowledge-base INDEX

Agent0 Site Overhaul — "Build Your Brain" Pivot

• Repositioned value prop: Claude Code + Obsidian = second brain (was "ops agent templates")
• Built "Build Your Brain" free interactive vault configurator (/build) with 6 role presets
• Removed Agent Studio ($3) entirely — new funnel: Free → $89 → Custom
• Built /insights journal page with 3 agent-perspective articles
• Built /about page ("How This Brain Works") with agent cards and operating principles
• Added Human/Agent language switcher + /api/agent-view markdown API
• Added AEO foundation: JSON-LD schemas, robots.txt inviting AI crawlers, sitemap with agent URLs
• Replaced fake SVG logos with official Claude + Obsidian logos (Wikimedia CC0)
• Added animations, terminal mockup showing real vault integration, visual polish
• Rewrote How It Works as 4-step brain system loop
• De-slopped all copy across every page (killed AI buzzwords)
• Fixed mobile nav (was completely broken)

Heartbeat Agent Fix

• Root cause: agent was deleting inbox files → every PR had 31 deletions → /sync couldn't auto-merge → PRs piled up (16 stale)
• Fix: updated trigger to archive (not delete) inbox files, never modify protected files, never create in projects/
• Updated /sync skill with Heartbeat-aware merge rules + stale duplicate cleanup
• Closed 15 stale Heartbeat PRs

Agent0 Project Archived

• Original agent0 concept (idea scoring, micro-SaaS pipeline) archived to archive/projects/agent0/
• Active product is now agent0-markops-app ("Build Your Brain")

Workflow Improvement

• Feedback saved: always use parallel agents for independent tasks (was single-threading everything)

2026-03-29

First Product Designed — Agent0 MarkOps App

• Designed agent0.markops.ai with 3 revenue streams: Agent Studio ($3/agent), Agent0 Framework ($89), Custom Setup (consulting)
• Agent Studio: 10 pre-built agent templates, form + chat customization, .zip export
• Framework: sanitized MySecondBrain repo + guided setup + Telegram community
• Tech: Next.js + Supabase + Stripe + Vercel + Claude API
• Design spec approved and saved to artifacts/plans/
• Updated decisions/Pick First Product to Ship — decision made
• Created project folder: projects/agent0-markops-app/

Agent0 MarkOps App — SHIPPED

• Built and deployed agent0.markops.ai — first product live
• Phase 1: Project scaffolding from starter template + Supabase schema
• Phase 2: Landing page (hero, 10 agent cards, pricing grid, FAQ, footer)
• Phase 3: Stripe integration (4 products, webhook handler, credit system)
• Phase 4: Seeded 10 agent templates into Supabase + /studio template library
• Phase 5: Agent builder (form → AI customization → preview → .zip export)
• Phase 6: Framework sales page ($89 one-time)
• Phase 7: User dashboard (credits, agents, framework access)
• Deployed to Vercel at agent0.markops.ai with all env vars configured

Content Creation SOP + Workflow Documented

• Created operations/sops/Publishing Articles on MarkOps.md — 6-phase SOP covering brief → draft → visuals → Webflow publish → distribution → post-publish, with all lessons learned baked in
• Created operations/workflows/Content Creation Workflow.md — maps which marketing skills and agents power each step, agent ownership per phase, workflow diagram
• Documents the full pipeline proven on first article: content-strategy → copywriting → copy-editing → ai-seo → distribute → publish → social-content

Task Board Cleared — 8 Tasks Executed

• Cloned ECC reference repo to ~/workspace/ecc-reference/ (130+ skills as local lookup)
• Added context monitor hook (PostToolUse) — warns at 50/80/100 tool uses per session
• Upgraded verification-before-completion — added stub detection patterns + goal-backward verification
• Upgraded brainstorming — added assumptions mode (confidence-scored assumptions instead of Q&A)
• Created personal goals framework at personal/goals/Goals.md — 6 goals with quarterly scoring
• Evaluated Swarmclaw — superseded by current Claude Code setup, task closed
• Evaluated UI/UX Pro Max — decision: adapt 2 new skills + upgrade 3 CRO skills (future task)
• Created decision doc: "Pick First Product to Ship" — NFL Buddy vs RevOps Automation vs Flower Letters
• Saved memory: inbox processing must complete full pipeline (research → synthesize → task → delete → update kanban)

Inbox Processing Pipeline — Full Execution

• Processed 5 inbox items through complete pipeline: research → synthesize → create notes → create tasks → delete inbox → update kanban
• Created 5 knowledge-base reference docs (LightRAG, ECC, Awesome CC, GSD, UI/UX Pro Max)
• Created 5 tasks from actionable findings (context monitor hook, ECC clone, KB index, skill upgrades, UI/UX eval)
• Created knowledge-base/INDEX.md — fast lookup table for Claude
• Fixed Content Pipeline kanban (moved article to Published)
• Fixed Task Board kanban (added 5 new proposed tasks, moved article to Done)
• Fixed Content Calendar (replaced broken dataview queries with manual schedule — Dataview plugin not installed)
• Saved memory: inbox processing must complete full pipeline, not just acknowledge items

/publish Skill Upgraded — Webflow + SEO + AEO

• Rewrote /publish skill with full Webflow API workflow (site IDs, collection IDs, asset upload, CMS item creation, publish + republish)
• Learned: URL pattern is /insights/article/[slug] not /article/[slug]
• Learned: images >4MB rejected — must compress with sips before upload
• Learned: full-width images can't be set via API — manual Webflow editor step
• Learned: must publish CMS items AND republish site (two separate calls)
• Added SEO: contextual internal linking to related markops.ai articles
• Added AEO: FAQ section with 5 structured questions for LLM citation
• Added: Related Articles section with 5 links
• Saved Webflow config and publishing workflow to Claude memory
• First article published: "How I Built My Second Brain" at markops.ai/insights/article/how-i-built-my-second-brain

Content Pipeline and Distribution System

• Created Content Pipeline kanban board (artifacts/Content Pipeline.md) — tracks content from idea to published
• Created Task Board kanban (tasks/Task Board.md) — visualizes task lifecycle
• Created Content Calendar (artifacts/Content Calendar.md) — schedule view with dataview queries
• Created /distribute skill — creates platform-native versions (LinkedIn article, X thread, X post)
• Created /publish skill — pushes content via Postiz CLI to X and LinkedIn
• Installed Postiz CLI (v2.0.12) — connected X (@maciekecom) and LinkedIn (maciekmarchlewski)
• Installed Obsidian Kanban and Calendar plugins
• Added Gemini API key and Postiz API key to credentials.env

Vault Link Strengthening

• Added ## Connections wikilinks to 14 idea files, 2 daily notes, 2 decision logs, 2 artifact files
• Rewrote projects/agent0/Agent0 Home.md — fixed all broken wikilinks from old vault structure
• Updated operations/Operations.md — added Tech Stack and Key Documents sections
• Updated projects/Projects.md — replaced broken dataview with manual project list
• Updated ideas/Ideas.md — added Idea Clusters grouping (Agent Apps, Outreach, Infrastructure, Content, Personal)
• Updated artifacts/Artifacts.md — added Article Ideas dataview section
• Updated personal/Personal.md — added Key Context (family names, address, priorities)
• Updated operations/sops/Daily Vault Ritual.md — fixed [[Home]] → [[My Second Brain]]
• Added credentials.env to .gitignore — was being pushed to GitHub
• Total: 30+ files updated with cross-links, strengthening the vault graph

2026-03-28

Vault Thinking Tools (7 new commands from Vin's Codebook)

• /context — load full life/work state into any session
• /trace [topic] — track how an idea evolved across the vault over time
• /connect [A] [B] — find unexpected bridges between two topics via link graph
• /ghost [question] — answer as Maciek, grounded in vault voice and positions
• /challenge [topic] — pressure-test beliefs, find contradictions and weak assumptions
• /drift — surface recurring themes across unrelated notes (subconscious patterns)
• /emerge — find idea clusters coalescing into potential projects

Task Management + Governance Framework

• Created tasks/ folder with task lifecycle: proposed → approved → in_progress → done
• Created /tasks skill for reviewing, approving, and rejecting task proposals
• Created governance/ folder with 4 documents:
  • Approval Matrix — what auto-executes vs what needs Maciek's approval
  • Agent Permissions — per-agent CAN/CANNOT rules
  • Escalation Rules — when agents must stop and ask
  • Rejection Log — feedback loop so agents learn from rejections
• Merged Inbox Processor + Orchestrator into single Vault Heartbeat agent (every 2h) due to trigger limits
• Vault Heartbeat now does: inbox processing + vault scan + task proposals via PR
• All tasks require human approval (human-in-the-middle policy)
• Rejection feedback loop: rejected tasks → logged with reason → Claude memory updated → pattern never proposed again

Vault Automation Pipeline

• Connected vault to private GitHub repo (markopsai/MySecondBrain)
• Installed Obsidian Git plugin for auto-sync every 30 min
• Created /sync skill — pulls from GitHub, auto-merges safe agent PRs, pushes local changes
• Created 4 scheduled cloud agents:
  • Inbox Processor — every 2 hours, extracts + synthesizes inbox items via PR
  • Nightly Agent Entry — 10pm ET daily, writes agent journal entry via PR
  • Weekly Reflection — Sunday 8pm ET, distills weekly patterns via PR
  • Monthly Reflection — 1st of month, big picture review via PR

Skills Created

• /daily — upgraded to Human Entry + Agent Entry system (agent entry powered by Claude memory)
• /capture — full extraction + synthesis engine (URLs, screenshots, PDFs, ideas)
• /reflect — weekly + monthly reflections with life balance scoring
• /sync — git sync + auto-merge for agent PRs
• /standup — cross-domain briefing across all projects
• /tldr — end-of-session summary with memory update

Vault Structure

• Established folder architecture: inbox, daily, ideas, projects, decisions, operations, artifacts, personal, people, knowledge-base, archive
• Renamed all hub files from generic Home.md to descriptive names (e.g., Decisions.md, Daily Journal.md) for better Obsidian graph view
• Created decisions/ folder for tracking what was decided, why, and what changed
• Created operations/ folder with systems/, sops/, workflows/ subfolders
• Created artifacts/ folder with subfolders: briefs, articles, page-copy, presentations, spreadsheets, plans
• Created personal/ folder consolidating family, investments, health, goals
• Created knowledge-base/ for reference material Claude learns from
• Moved Starter App Template out of vault to ~/workspace/starter-app-template/
• Removed content/ folder — strategy docs belong in their project folder
• Removed imports/ folder — replaced by knowledge-base/

Memory & Learning Rules

• Enabled auto-memory: Claude updates memory files after every session
• Saved feedback: no dates in filenames (dates go in frontmatter)
• Saved feedback: strategy docs belong inside project folders, not generic folders
• Saved reference: starter app template location for building apps

SOPs & Workflows

• Created Daily Vault Ritual SOP — nightly process: /daily → /sync → /tldr → close laptop

Decisions

• Logged vault structure decisions (5 entries)
• Logged automation pipeline architecture decision
• Logged daily journal human + agent entry decision
• Logged automated reflections decision

2026-03-27

Initial Setup

• Created MySecondBrain vault
• Installed obsidian-skills (defuddle, obsidian-markdown, obsidian-bases, obsidian-cli, json-canvas)
• Migrated content from old "Maciek's Brain" Obsidian vault
• Created CLAUDE.md with vault context and rules
• Set up initial agent team configuration in RevOpsAi project