Link-Based Plans Dashboard

Someone showed Brandon a method. Here is each one in plain words, scored, with the one tiny step to make it real.

These are NOT the normal YouTube transcript jobs. These are the X links, YouTube methods, plugin ideas, Claude Code tricks, Obsidian workflows, and n8n automations someone showed Brandon and he wanted turned into a usable workflow. The score says if it is real yet.

25plans + methods
2green
9yellow
14red

Claude Code adoption plans (9)

Claude Code "secret codes" / hotkeys YELLOW

What it is
A plan to start using hidden Claude Code commands like ultrathink, /insights, /btw, /context, and /compact, plus get one sound ping when a session finishes.
Problem it solves
Brandon was not using the power commands that make Claude think harder, save tokens, and tell him when a job is done.
What Brandon wanted
To learn the secret commands from a YouTube video and actually use them, and get a beep when Claude finishes so he can walk away.
Connects to
/Users/drwu/Projects/_outputs/claude-code-adoption/, /Users/drwu/.claude/settings.json, /Users/drwu/.claude/projects/-Users-drwu/memory/, session-audit skill
Using it?
Partly. Half right. The cheat-sheet got written and two memory notes about ultrathink and shortcuts exist, so the knowledge is saved. But the plan claims the notification hook is "already satisfied," and that hook was in settings.json 17 days BEFORE the plan, so Phase 2 did no new work. The ultrathink and clear-between-tasks habit is just a habit, with no way to prove it is being used.
Why this color
More than just an idea: the plan AND a derivative cheat-sheet exist, plus two matching memory files. But no NEW machinery was wired by this plan. The one notification hook it points to predates it by 17 days, only 1 of 13 plans was ever enqueued, and the ultrathink habit cannot be verified as in use. Written and partly mapped, not live and producing work.
One tiny first step
Open /Users/drwu/.claude/_state/plans_enqueued.json and add plan-4 so the /plans board stops showing it as never-started, then run one real planning prompt with the word ultrathink in it.
Checked on disk
/Users/drwu/Projects/_outputs/claude-code-adoption/claude-secret-codes-plan.md exists (5494 bytes, created 2026-06-06)
/Users/drwu/Projects/_outputs/claude-code-adoption/commands-cheatsheet.md exists (3889 bytes, created 2026-06-08), the Phase 1 deliverable
plans_enqueued.json contains only one UNRELATED plan (avoidant-narcissist-ai-video-pipeline-plan.md); plan-4 / claude-secret-codes is NOT enqueued
Stop hook 'curl -H Title: Claude done ... ntfy.sh/brandonmac' is present in /Users/drwu/.claude/settings.json AND in settings.json.bak-pre-conductor-2026-05-20, proving the Phase 2 hook predates the 2026-06-06 plan by 17 days
The cheatsheet itself states Phase 2 is 'already satisfied' and 'No new hook needed'
memory files reference_ultrathink_keyword.md and reference_claude_code_shortcuts.md exist but are sourced from a '@0xCodez article' not the plan's two videos, and share originSessionId 0d98aa9f, so they are independent of this plan
source transcript ~/New Desktop/yt-claude-code/transcripts/rabGqnyd_Zw.txt exists (11286 bytes), so the plan provenance is real
Brandon-Vault session note 2026-06-06-1551 only records WRITING the batch of adoption plans, not acting on them

Official Todoist connector RED

What it is
A written plan to add Claude's official Todoist connector so Claude can both add tasks to Todoist and read them back. It came from two YouTube videos Brandon watched.
Problem it solves
Right now Claude can only push tasks one way and needs a secret token in a file. Brandon wants Claude to also read his task list back, like for an end-of-day review, using a safe login instead of a token.
What Brandon wanted
He wanted the safe, official login-based connector turned on so he could capture tasks from his phone and ask Claude things like "which of today's tasks help my goals," then decide if it replaces his old token setup.
Connects to
~/.claude/skills/chat-conductor/_lib/todoist_sync.py (existing token-based push path the plan would replace), Brandon-Vault/40 Systems/App Integrations/APP_STACK.md (where Phase 3 decision was supposed to be written), Brandon-Vault/Inbox/todoist-queue.jsonl (the offline fallback queue the plan mentions), ~/.claude/settings.json (where a connector would be wired)
Using it?
No. Not using it at all. The plan got written but no step was done. There is no Todoist connector in settings.json, the auth cache only lists Adobe and Atlan, and every Todoist trace on disk is still the OLD token-and-REST path (todoist_sync.py calling api.todoist.com with TODOIST_TOKEN), which is the exact thing this plan said it would replace. APP_STACK.md still describes Todoist as the token-and-.env path with no Phase 3 decision note. The plans board confirms it: only 1 unrelated plan was ever enqueued, and this one was not.
Why this color
Red because it is only a written plan with zero action on disk. No connector exists in settings.json or the auth cache, no Phase 3 decision was written into APP_STACK.md, no session note or artifact shows the connector being added or the three workflows being tested, and the plans_enqueued.json file proves this plan was never enqueued.
One tiny first step
Open Claude Settings, then Connectors, click Add, and pick Todoist, then log in. That one click finishes Phase 1 step 1 and proves the whole plan is real.
Checked on disk
/Users/drwu/Projects/_outputs/claude-code-adoption/todoist-official-connector-plan.md exists and is fully written (3 phases, sourced from videos UVV29TKjXTM + axejg_WIKxc)
/Users/drwu/.claude/_state/plans_enqueued.json contains only the avoidant-narcissist video plan; grep for 'todoist' returns nothing -> this plan was never enqueued
settings.json top-level keys are [permissions, model, hooks, statusLine, autoUpdatesChannel, skipWorkflowUsageWarning, theme, remoteControlAtStartup, inputNeededNotifEnabled, agentPushNotifEnabled] -- no connectors / mcpServers / enabledMcpjsonServers entry
/Users/drwu/.claude/mcp-needs-auth-cache.json lists only 'Adobe for creativity' and 'atlan' -- no Todoist
All Todoist hits in ~/.claude are the pre-existing chat-conductor sync; todoist_sync.py uses REST api.todoist.com with TODOIST_TOKEN (lines 57,63,77,91,180-181), the token path the plan intended to replace
Brandon-Vault/40 Systems/App Integrations/APP_STACK.md still describes Todoist as 'login-based (app session + REST API token in .env)' with no official-connector decision note (Phase 3 deliverable missing)
grep across Brandon-Vault for 'official todoist connector' / 'add from todoist' / 'todoist mcp url' returned no matches -- no test or workflow evidence
todoist-queue.jsonl not found at either Desktop or New Desktop Brandon-Vault/Inbox path

Obsidian + Claude in the terminal RED

What it is
A plan to put a little terminal window right inside Obsidian, then start Claude Code from there so Brandon's notes and his coding helper live in one screen.
Problem it solves
Right now Brandon opens a separate terminal to run Claude against his vault. This plan would let him see his notes, his graph, and Claude side by side, with no jumping between windows.
What Brandon wanted
Brandon watched two YouTube videos and wanted to copy the trick: dock a terminal in Obsidian, launch Claude at the vault root, and add a small vault CLAUDE.md so Claude knows it is in his notes folder.
Connects to
/Users/drwu/Desktop/Brandon-Vault/, /Users/drwu/Desktop/Brandon-Vault/README.md, /Users/drwu/Desktop/Brandon-Vault/_Dashboards/ (Conductor Hub, Workflow Command Center), /Users/drwu/Projects/_outputs/claude-code-adoption/, /Users/drwu/New Desktop/yt-claude-code/transcripts/B35SWx_4BNM.txt
Using it?
No. Not using it. The plan was written but never done. Phase 1 wants a terminal plugin in Obsidian, and it is not installed. The 13 enabled plugins in community-plugins.json have no terminal plugin. Phase 2 wants a small CLAUDE.md at the Brandon-Vault root, and that file does not exist. The /plans board shows this plan was never even enqueued, so no step was ever started.
Why this color
It is only the written plan with zero action on disk. No terminal plugin is installed, no vault CLAUDE.md exists, and the enqueue file holds a different plan, so none of the three phases were started.
One tiny first step
Open Obsidian, go to Community plugins, search "terminal", install and enable the terminal plugin. That is Phase 1 step 1 and it makes the rest possible.
Checked on disk
Plan file exists at /Users/drwu/Projects/_outputs/claude-code-adoption/obsidian-claude-terminal-plan.md (3 phases, status: active, created 2026-06-06)
No vault-scoped CLAUDE.md: 'NO CLAUDE.md at Brandon-Vault root' (checked /Users/drwu/Desktop/Brandon-Vault/CLAUDE.md and /Users/drwu/New Desktop/Brandon-Vault/CLAUDE.md) — Phase 2 deliverable missing
No terminal plugin: .obsidian/community-plugins.json lists 13 plugins (dataview, templater, tasks, quickadd, todoist-sync, periodic-notes, excalidraw, excalibrain, local-gpt, smart-connections, readwise, citation, local-rest-api) and none is a terminal plugin; find for *terminal* returned nothing — Phase 1 deliverable missing
/Users/drwu/.claude/_state/plans_enqueued.json contains only the avoidant-narcissist video pipeline plan, NOT this Obsidian terminal plan — confirms it was never enqueued
No 'terminal' reference in /Users/drwu/.claude/settings.json or /Users/drwu/.claude/hooks; no obsidian/terminal skill under /Users/drwu/.claude/skills
Vault grep hits for B35SWx_4BNM were an unrelated YouTube transcript folder and a session note, not implementations of this plan

Plan-mode discipline RED

What it is
A short written plan that says: when a Claude Code job has 3 or more steps, let Claude look at the code in read-only "plan mode" first and show a plan before it changes any files. Brandon got the idea from two YouTube videos.
Problem it solves
Brandon runs lots of big multi-step jobs (book pipelines, the shift queue, MIDI loops). Sometimes Claude jumps in and edits or deletes things before Brandon can check the plan. This idea makes Claude show its plan first so it does not break a rule like "never delete" or "never write to Desktop."
What Brandon wanted
He wanted a house rule, and maybe a little reminder hook, so multi-phase work always opens in plan mode and gets checked against his safety rules before any file gets touched.
Connects to
/Users/drwu/Projects/_outputs/claude-code-adoption/, /Users/drwu/.claude/settings.json, /Users/drwu/.claude/hooks/honesty_gate_nudge.py, /Users/drwu/CLAUDE.md, /Users/drwu/.claude/_state/plans_enqueued.json, /Users/drwu/New Desktop/yt-claude-code/transcripts/
Using it?
No. It is not being used. The plan lists its own two jobs and neither one got done. Phase 1 said to write a note called plan-mode-first.md in the claude-code-adoption folder. That file does not exist anywhere. Phase 2 said to describe a reminder hook and leave it unwired for approval. No plan-mode hook script exists in ~/.claude/hooks/, and settings.json has zero mention of plan mode. There is a separate "three-phase plan" rule in CLAUDE.md and the honesty_gate_nudge.py hook, but that is an older rule about writing a plan in chat. It is not the read-only plan-mode discipline this plan describes, so it does not count as this plan being live.
Why this color
This is just the written plan with no action taken. The plan's own Phase 1 file (plan-mode-first.md) was never created, no plan-mode hook exists, settings.json has no plan-mode wiring, and the plan was never enqueued into the shift queue (plans_enqueued.json holds only one unrelated plan). A Vault session note from the day it was written confirms the author "did NOT enqueue anything" and "did NOT edit any live file."
One tiny first step
Write the one file Phase 1 asks for: /Users/drwu/Projects/_outputs/claude-code-adoption/plan-mode-first.md, with the rule (3+ phases = open in plan mode, check against never-delete and no-Desktop-writes first) and the quoted read-only line from the gbetp6D7J_Q transcript.
Checked on disk
Read /Users/drwu/Projects/_outputs/claude-code-adoption/plan-mode-discipline-plan.md (3462 bytes, created 2026-06-06, status: active, phases: 2)
ls of /Users/drwu/Projects/_outputs/claude-code-adoption/ shows 13 *-plan.md files but NO plan-mode-first.md (the Phase 1 deliverable)
find for plan-mode-first* across ~/Projects and Brandon-Vault returned nothing
grep of settings.json and settings.local.json for plan-mode/plan_mode/planmode = no matches
No plan hook in /Users/drwu/.claude/hooks/ and no /Users/drwu/.claude/skills/*plan* folder
/Users/drwu/.claude/_state/plans_enqueued.json contains only one entry (avoidant-narcissist-ai-video-pipeline-plan.md), NOT this plan
honesty_gate_nudge.py mentions a 'three-phase plan' but that is the older CLAUDE.md chat-plan rule, not read-only plan-mode; CLAUDE.md has 2 three-phase references unrelated to this plan
Vault session note 2026-06-06-1550...fb8fcfd5.md states the authoring session 'Did NOT enqueue anything' and bridge reported '1 enqueued | 13 active workflow plans NOT enqueued'
Both source transcripts exist and are real: gbetp6D7J_Q.txt (2219 bytes) and 5Pcsqx-CbQY.txt (29527 bytes)

Sub-agents + worktrees in parallel RED

What it is
A written plan that copies tricks from a YouTube video so Claude can run many helper bots at once, work on safe side branches, and block dangerous delete commands.
Problem it solves
Brandon wants Claude to do long jobs by itself without watching it, and to never delete his files by accident.
What Brandon wanted
Cheap helper bots (Haiku) doing the boring reading, the main brain (Opus) doing the thinking, a safety rule that blocks rm and trash, and to-do steps that double-check their own work.
Connects to
/Users/drwu/Projects/_outputs/claude-code-adoption/, /Users/drwu/.claude/settings.json, /Users/drwu/.claude/settings.local.json, /Users/drwu/.claude/_state/plans_enqueued.json, /Users/drwu/Desktop/Brandon-Vault/Sessions/Recents-7onething1-2026-06-06/
Using it?
No. No. The plan was written but never turned on. Worse, the real settings do the OPPOSITE of Phase 2: settings.json allows Bash(*) (everything) and settings.local.json has 187 allow rules including Bash(rm -rf *) with ZERO deny rules, so the never-delete safety the plan asked for is not enforced at all.
Why this color
Only the plan document exists. None of its three phases were built: the Phase 1 routing note is missing, the Phase 2 deny list is missing (and the live settings allow rm -rf, the reverse of the plan), and no skill or hook adds self-checking to-dos. The plans board confirms it was never enqueued.
One tiny first step
Add one deny line to ~/.claude/settings.json permissions that blocks Bash(rm:*) and Bash(git rm:*). That single Phase 2 step makes the plan real and matches the never-delete gate.
Checked on disk
Plan file exists: /Users/drwu/Projects/_outputs/claude-code-adoption/sub-agents-worktrees-parallel-plan.md (5264 bytes, created 2026-06-06)
Phase 1 deliverable MISSING: subagent-routing.md does not exist in the claude-code-adoption folder (ls confirmed 13 plan docs, none named subagent-routing.md)
Phase 2 NOT implemented and reversed: settings.json permissions = allow ['Bash(launchctl load:*)','Bash(launchctl unload:*)','Bash(*)'], zero deny rules
settings.local.json has 187 allow rules and 0 deny rules, including 'Bash(rm -rf *)' and 'Bash(rm -rf ~/.cache/*)' — the never-delete gate is NOT enforced at the permission layer
Phase 3 NOT implemented: grep of ~/.claude/skills and ~/.claude/hooks for subagent-routing/worktree/self-checking-todo returned no matches
plans_enqueued.json contains only one entry, the avoidant-narcissist video pipeline plan, NOT this plan — so it was never enqueued
Brandon-Vault session note (2026-06-06-1551...7e19ba5c.md) lists this plan among docs 'produced by the parallel sessions' and says the tracker 'flags each active plan never queued as STUCK'

Standardize connectors + custom instructions RED

What it is
A written to-do plan that came from two YouTube videos. The idea is to make one clear list of which app connector (like Gmail or Google Drive) does which job, plus a few short "always-on" instruction notes for Claude, then prove it works once.
Problem it solves
Brandon has lots of connectors wired up (Gmail, Drive, Notion, Calendar, Todoist) but no map of which one to use for which job. This plan was supposed to write that map so research questions pull from the right place every time.
What Brandon wanted
He wanted a tidy standard: one connector-to-job map, three reusable briefing notes (book pipeline, music, vault), and one real research-mode test to show it actually works.
Connects to
/Users/drwu/Projects/_outputs/claude-code-adoption/, /Users/drwu/Desktop/Brandon-Vault/40 Systems/App Integrations/APP_STACK.md, /Users/drwu/Desktop/Brandon-Vault/_Dashboards/Connectors Index.md, /Users/drwu/.claude/skills/YTtranscript-researchmode/
Using it?
No. Not using it. The plan names three things it should make: a connector_job_map.md file, three briefing templates, and one recorded research-mode query. None of the three exist on disk. The only file that even says "connector_job_map" is the plan itself. So nothing was built and nothing is running.
Why this color
Only the written plan exists. None of its three promised outputs (connector_job_map.md, three briefing templates, one recorded research-mode query) are anywhere on disk, settings.json has no connector or custom-instruction wiring, and the plans board shows this plan was never enqueued. Pure idea, no action.
One tiny first step
Do Phase 1 only: make the one file the plan promised. Create /Users/drwu/Projects/_outputs/claude-code-adoption/connector_job_map.md and list each connector next to the job it does, using APP_STACK.md as the source. That one file makes the plan real.
Checked on disk
Plan read at /Users/drwu/Projects/_outputs/claude-code-adoption/claude-connectors-custom-instructions-plan.md; Phase 1 promises connector_job_map.md, Phase 2 promises three briefings, Phase 3 promises one recorded research-mode query
find /Users/drwu -name connector_job_map.md returned NONE
Output folder /Users/drwu/Projects/_outputs/claude-code-adoption/ holds only 13 .md plan/cheatsheet files, no map and no briefing templates
grep for connector_job_map across Projects, Brandon-Vault, and .claude matched only the plan file itself
grep settings.json for connector|custom-instruction|research-mode = no matches
No skill dir named connector/briefing/custom-instr/research-mode under ~/.claude/skills/
Brandon-Vault/Connectors/ is empty and dated 2026-05-20, predating the 2026-06-06 plan
Brandon-Vault/_Dashboards/Connectors Index.md is conductor-auto-generated 2026-05-28 (predates plan) and itself says 'no MCP servers configured in settings.json'
~/.claude/_state/plans_enqueued.json contains only the avoidant-narcissist-ai-video-pipeline plan; this plan was never enqueued
No briefing/custom-instruction file dated after 2026-06-06 anywhere; YTtranscript-researchmode skill exists but does not reference this plan

PRD-first Claude to n8n build RED

What it is
A note Brandon wrote about a trick from some YouTube videos: write a short "what I want" doc (a PRD) first, then have Claude build the n8n robot-workflow one piece at a time, fixing each piece before going on.
Problem it solves
Right now Brandon's n8n Workflow Hub has no set way to turn a new workflow idea into an actual built workflow. Ideas get described in words but there is no step that makes Claude build them step by step.
What Brandon wanted
He wanted a reusable habit: write a PRD template, write down the build-and-debug loop, then prove it works by building one small real workflow with it.
Connects to
/Users/drwu/Desktop/Brandon-Vault/40 Systems/n8n Workflows/ (idea_miner.py, n8n_daily_idea_mining_workflow.json, install_workflow_hub.sh), /Users/drwu/Projects/_outputs/claude-code-adoption/ (the folder of 13 adoption plans), /Users/drwu/New Desktop/yt-claude-code/transcripts/ (the source videos)
Using it?
No. Not using it. The plan has three phases and zero of them were done. Phase 1 says make a file called n8n_prd_template.md, that file does not exist anywhere on disk. Phase 2's build-loop doc was never written. Phase 3's "steering log" from a real test build was never recorded. The plan is just the plan.
Why this color
It is only the written plan with no action taken. Every artifact the plan promises is missing, and the plans board confirms it was never enqueued.
One tiny first step
Do Phase 1 only: create one file, "/Users/drwu/Desktop/Brandon-Vault/40 Systems/n8n Workflows/n8n_prd_template.md", with the trigger / inputs / per-node steps / APIs / expected output / success-check fields, and fill it in once using the existing idea_miner.py workflow as the worked example.
Checked on disk
Plan file exists at /Users/drwu/Projects/_outputs/claude-code-adoption/n8n-from-claude-prd-first-plan.md, created 2026-06-06, status: active, 3 phases
/Users/drwu/.claude/_state/plans_enqueued.json contains ONLY the avoidant-narcissist video plan, NOT this n8n plan, so it was never enqueued
find for *prd* across Brandon-Vault and ~/.claude returns nothing; n8n_prd_template.md does NOT exist (the Phase 1 deliverable)
ls of Brandon-Vault/40 Systems/n8n Workflows/ shows only pre-existing files (idea_miner.py, n8n_daily_idea_mining_workflow.json, install_workflow_hub.sh, N8N_IDEA_MINING_SPEC.md) - none added by this plan
No n8n skill dir in ~/.claude/skills (glob *n8n* = no matches); no n8n/prd reference in ~/.claude/settings.json or any file in ~/.claude/hooks/
grep for 'prd-first'/'node-by-node'/'steering log' across the vault hit only the 2026-06-06-1550 session transcript, which is the very session that AUTHORED the doc and states '1 enqueued | 13 active workflow plans NOT enqueued' and 'Did NOT enqueue anything'

Route community skills into n8n RED

What it is
A written plan to teach Claude Code how to build n8n automation workflows. It would add an n8n "MCP" tool plus a bundle of n8n "skills" (cheat-sheet notes on n8n logic) copied from GitHub, so Claude can write good workflows instead of guessing.
Problem it solves
Brandon runs n8n on his Mac and writes its workflow files by hand, which is slow and easy to get wrong. This plan would let Claude build and fix those workflows for him, the way the YouTube video showed (a form to a PDF reader to Google Sheets).
What Brandon wanted
He watched two YouTube videos (TztJaN8ep60 and lGLwLOERp6c) and wanted Claude to build his n8n workflows using a skills layer that knows n8n's expressions, not just plain node names. He wanted to compare this community route against the official n8n MCP plan and pick one.
Connects to
/Users/drwu/Projects/_outputs/claude-code-adoption/n8n-community-skills-route-plan.md, /Users/drwu/Desktop/Brandon-Vault/40 Systems/n8n Workflows/, /Users/drwu/Desktop/Brandon-Vault/40 Systems/n8n Workflows/n8n_daily_idea_mining_workflow.json, /Users/drwu/Projects/_outputs/claude-code-adoption/n8n-official-mcp-server-plan.md, com.drwu.n8n launchd job
Using it?
No. No. The plan was written on Jun 6 and never touched again. Nothing in it got built. There is no n8n MCP in ~/.claude/settings.json (no mcpServers key at all), no n8n skill folder in ~/.claude/skills/, no cloned n8n-mcp or n8n-skills repo on disk, no workspace CLAUDE.md for an n8n build role, and no Phase 3 standing-rule note in the n8n Workflows folder. The plans board confirms it: plans_enqueued.json holds only 1 plan, and it is the avoidant-narcissist video one, not this. The thing the plan builds ON is real (the com.drwu.n8n launchd job runs, PID 3509, and hand-written workflow specs exist), but the route itself was never wired in.
Why this color
It is only the written plan with zero action. No MCP connected, no skills cloned, no note created, never enqueued. The n8n instance it relies on is real, but the route in this plan was never started.
One tiny first step
Open the plan, do just Phase 1 step 1: read the sibling file n8n-official-mcp-server-plan.md and write one line at the top of the plan saying "official MCP first, community skills as add-on" or the reverse. That one decision unblocks everything else.
Checked on disk
plans_enqueued.json contains only 1 entry (avoidant-narcissist-ai-video-pipeline-plan.md), NOT this n8n plan
~/.claude/settings.json has NO mcpServers/mcp key, so no n8n MCP is connected
ls ~/.claude/skills/*n8n* returns no matches, no community n8n skill folder
find found no cloned n8n-mcp or n8n-skills repo under ~/Projects or ~/.claude
No standing-rule note in 'Brandon-Vault/40 Systems/n8n Workflows/' (grep for 'skills layer/community skills/idiom layer' found nothing)
No workspace CLAUDE.md describing an n8n build role
No n8n reference in ~/.claude/hooks/
Plan file created and modified at the same timestamp Jun 6 14:41:56, never edited after writing
Premise IS real: launchctl shows com.drwu.n8n running (PID 3509) and idea_miner.py + n8n_daily_idea_mining_workflow.json exist in the n8n Workflows folder

Prompt superpowers skill (XML tags + prefill + constraint stacking) RED

What it is
A written plan to turn a YouTube prompting trick into a reusable Claude skill. The trick is three parts: XML tags to shape the input, "prefill" to start the answer with no chit-chat, and stacking rules to control quality.
Problem it solves
Brandon runs lots of jobs that need the same exact format with no fluff, like MIDI critiques, BandBook cards, devotional chapters, and Etsy copy. This plan would give him one template so every bulk job comes out clean and on-format the first time.
What Brandon wanted
A shared prompt template (four labeled blocks: instructions, context, task, output format) plus a prefill move and a constraint checklist, so his generator skills could all point at one good pattern instead of each guessing.
Connects to
~/.claude/skills/_shared/ (where the template was supposed to live), ~/Desktop/project-audit-dashboard/voice-models/music_writer_voice_prompt.md (the voice ruleset it would cross-link), the midi-16-runs-* generator skills and BandBook card pipeline (the bulk jobs it would serve), ~/New Desktop/yt-claude-code/transcripts/jWyb488osEY.txt (the source video transcript, which is real)
Using it?
No. Not using it at all. The plan is just a markdown file. None of the three phases happened on disk: no template file, no skill, no hook, no wiring. So there is no "right or wrong use" yet, there is nothing built.
Why this color
The plan was written on 2026-06-06 but never acted on. I checked disk directly and found no template in _shared/, no prompt-superpowers skill folder, no hook, no mention in settings.json, and no reference to the video id anywhere in the skills tree or the scoped Vault. The plans_enqueued.json board lists only ONE plan (the avoidant/narcissist video pipeline), not this one, so it was never even queued. It is a written plan with zero action, which is the definition of red.
One tiny first step
Do Phase 1 only: create one file at ~/.claude/skills/_shared/prompt_superpowers.md with the four XML blocks and ONE real worked example (a BandBook chart card), and a link to the voice ruleset. That alone turns red into yellow.
Checked on disk
Plan file exists and is real: /Users/drwu/Projects/_outputs/claude-code-adoption/prompt-superpowers-skill-plan.md (4276 bytes, one of 13 plan .md files in that folder)
Source transcript exists: /Users/drwu/New Desktop/yt-claude-code/transcripts/jWyb488osEY.txt (22579 bytes)
NO prompt_superpowers file in ~/.claude/skills/_shared/ (ls + find both returned nothing)
NO prompt-superpowers skill folder under ~/.claude/skills/ (ls | grep prompt = none)
NO mention of 'superpower', 'prompt_superpower', or constraint-stack in ~/.claude/settings.json; the hooks list has 20+ hooks, none of them this one
grep of ~/.claude/skills *.md for 'jWyb488osEY' / 'prompt-superpower' returned zero hits
~/.claude/_state/plans_enqueued.json contains only one entry: the avoidant-narcissist-ai-video-pipeline-plan, NOT this plan
Scoped grep of Brandon-Vault 20 Projects + 40 Systems + _Dashboards for the plan/video id = zero hits

Project plans (3)

Avoidant / Narcissist AI-Video Pipeline YELLOW

What it is
A plan to make short AI-narrated videos about avoidant and narcissist relationship patterns, with a cloned voice reading scripts over visuals.
Problem it solves
Brandon writes strong relationship teaching but it sits in text. Videos reach people who never read, and they can post daily without filming anyone.
What Brandon wanted
A repeatable machine: script in, finished narrated video out, good enough to post.
Connects to
~/Desktop/Brandon-Vault/20 Projects/avoidant-narcissist-ai-video-pipeline-plan.md, ~/Projects/_outputs/avoidant-video-pipeline/, ElevenLabs narration (Brandon clone rejected, Lester broken)
Using it?
Partly. The build is real: a format spec, a decisions doc, two rounds of test renders and test videos. The right voice took weeks to settle. No video has shipped to a channel.
Why this color
Lots of working pieces on disk (specs, test renders, phase-1 status from Jun 7) and zero published videos. Started for real, not yet producing.
One tiny first step
Pick the best clip in test-renders-v2 and post that one video to one channel. Shipping one beats polishing ten.
Checked on disk
avoidant-narcissist-ai-video-pipeline-plan.md (37KB, Jun 4)
avoidant-video-phase1-status.md (Jun 7)
avoidant-video-DECISIONS-2026-06-06.md
~/Projects/_outputs/avoidant-video-pipeline/{test-renders,test-renders-v2,test-videos}

Ten-Substack Automation YELLOW

What it is
A plan to run many Substack newsletters at once (tarot, Christian, music, relationships) with Claude writing drafts on a schedule.
Problem it solves
One person cannot hand-write that many newsletters. The plan turns finished book and corpus work into posts so each lane publishes weekly without fresh writing.
What Brandon wanted
Each content lane earning on its own, fed by work he already produced.
Connects to
~/Desktop/Brandon-Vault/20 Projects/ten-substack-automation-plan.md, ~/Projects/_outputs/substack/ (auth, census, pipeline, CEO_10_PHASE_LAUNCH_PLAN.md), substack-transfer-operator skill, brandonpag3 Substack (created 2026-06-10)
Using it?
Partly. The machinery is half-built: a launch plan, an auth folder, a census of content, a pipeline folder, and one lane's launch post already drafted. One account exists, three lanes have no account yet.
Why this color
Real outputs exist in ~/Projects/_outputs/substack and one account is live, and no lane publishes on a schedule yet.
One tiny first step
Publish the already-written lane3_launch_post.md to the brandonpag3 Substack. It is sitting there finished.
Checked on disk
ten-substack-automation-plan.md (32KB)
~/Projects/_outputs/substack/CEO_10_PHASE_LAUNCH_PLAN.md
~/Projects/_outputs/substack/lane3_launch_post.md
memory: brandonpag3 created 2026-06-10, 3 lanes TBD

Kate + Chiron Relational Engine RED

What it is
A plan to turn two tarot-reader "voice" robots (Kate and Chiron) into a machine that writes whole tarot readings about relationships, then locks them so nobody can copy the secret recipe.
Problem it solves
Brandon wants to sell tarot readings without giving away how he makes them. The engine should write readings in Kate's caring, psychology-style voice and add real relationship advice, so he has a product to sell on Etsy and Substack that others cannot steal.
What Brandon wanted
A finished, protected reading-maker: type in some cards and a question, get back a polished Kate or Chiron reading (or even a video), all owned only by him. Three parts: lock the voices, build a relationship-diagnosis mode, and add copy-protection so buyers get the reading but never the recipe.
Connects to
/Users/drwu/Desktop/Brandon-Vault/20 Projects/kate-chiron-relational-engine-plan.md, /Users/drwu/Desktop/project-audit-dashboard/voice-models/, /Users/drwu/Projects/_outputs/kate-engine-v3/, /Users/drwu/.claude/skills/tarot-chapter-voice-video/, /Users/drwu/.claude/skills/tarot-canon/, /Users/drwu/.claude/skills/tarot-guide-builder/, /Users/drwu/.claude/skills/relationship-psych-canon/
Using it?
Partly. Only the very first piece is being used. The voice generator and three judge scripts already exist and run, and there is one real Kate run on disk (kate-engine-v3, June 6: a 5-card Empress-reversed reading taken through six versions plus a report). That is voice tuning, not the relational engine. None of the new build steps the plan calls for got made.
Why this color
The plan is detailed and the base voice tools already exist, but almost none of the plan's own deliverables shipped. Every new file the plan names is missing on disk, and even the tiny first step (a Kate row in the rubric) was never done. What does exist (kate-engine-v3) is older voice tuning, not the relational engine. That is "only an idea / half-plan" territory, so red.
One tiny first step
Do the single smallest Phase 1-A step: add the missing Kate row to Section 7 (Emotional Register Match) in /Users/drwu/Desktop/project-audit-dashboard/voice-models/voice_validation_rubric.md. Right now that section only lists Chiron, m3ksmith, and Eyephim.
Checked on disk
Plan file exists and is detailed: /Users/drwu/Desktop/Brandon-Vault/20 Projects/kate-chiron-relational-engine-plan.md (status: active, created 2026-06-04)
MISSING relational_card_map.json (Phase 2-A) in voice-models/
MISSING validate_card_map.py (Phase 2-A)
MISSING relational_templates/ folder (Phase 2-C)
MISSING batch_generate.py (Phase 1-C)
MISSING provenance_log.jsonl and provenance_check.py (Phase 3-B)
MISSING /Users/drwu/Desktop/Brandon-Vault/40 Systems/Tarot Engine License.md (Phase 3-C)
MISSING [TAROT] Kate Voice V2 LOCKED.md in iCloud SVG/ (Phase 1-A); only unlocked [TAROT] Kate Voice V2.md is there
voice_generate.py has only --reader/--querent/--question/--cards/--out; NO --mode, --to-mp4, --watermark, or --attachment flags the plan scopes
voice_validation_rubric.md Section 7 (lines 77-80) still lists only Chiron, m3ksmith, Eyephim; no Kate row (Phase 1-A's literal first step undone)
Only real output: /Users/drwu/Projects/_outputs/kate-engine-v3/ (June 6) = one 5-card Empress-reversed Kate reading across v2-v6 + kate_engine_v3_report.md; voice-tuning work, not the relational engine
drwu-htmls/public has chiron-tarot, chiron-matilda-cards, kate-combo-fraud-eval = separate older tarot work, not this engine

n8n automation (Innate In) (1)

n8n automation workflows (6 written specs) GREEN

What it is
Six little robot assembly lines written as JSON files: morning briefing, sales monitor, handoff arrival, book pipeline, research drop-zone, voice-rule batch. n8n is the app that runs them.
Problem it solves
Moving files, checking sales, and filing research by hand eats time every day. These robots do the moving on rules.
What Brandon wanted
The boring daily plumbing to run itself.
Connects to
~/Desktop/Brandon-Vault/_n8n-workflows/ (6 JSON specs), ~/my_n8n (install), ~/.n8n/database.sqlite (18 workflows), launchd com.drwu.n8n
Using it?
Yes. Mostly yes, and better than anyone assumed: n8n runs as a background service and 5 of the 6 written specs are imported AND switched on (sales monitor, handoff arrival, book pipeline, research drop-zone, voice-rule batch). The morning briefing spec was never imported, and 8 Christian-book workflows sit imported but switched off.
Why this color
The service is alive right now (launchd job com.drwu.n8n, PID 3509) and five robots are active in the database. This is the rare plan that actually shipped.
One tiny first step
Import 04-morning-briefing.json into n8n and flip it on. Five minutes, finishes the set.
Checked on disk
launchctl list shows com.drwu.n8n running (PID 3509)
sqlite ~/.n8n/database.sqlite: 18 workflows, 05/06/07/08/09 specs active=1
04-morning-briefing.json present in vault, absent from n8n DB

Saved YouTube / X method themes (12)

Claude Code memory / RAG / 3-file system GREEN

What it is
Videos comparing ways to give Claude a lasting memory: memory files, RAG lookups, and simple 3-file systems that any model can read.
Problem it solves
Without memory, every chat starts from zero and repeats old mistakes. A memory system makes lessons stick.
What Brandon wanted
Claude remembering his rules, projects, and past failures across every session.
Connects to
~/.claude/projects/-Users-drwu/memory/ (366 files), MEMORY.md index loaded every session, hard-gate hooks that enforce remembered rules
Using it?
Yes. Yes, and it is the strongest method on this page: 366 memory files, an index loaded into every chat, and hooks that turn remembered feedback into enforced gates. One real flaw: the index is over its size limit, so part of it gets cut off each session.
Why this color
In daily use and visibly steering work. The overflow trim is maintenance, not adoption.
One tiny first step
Trim MEMORY.md under its 24.4KB limit (it is 26.7KB now, so the tail entries silently drop). One editing pass fixes it.
Checked on disk
memory/ holds 366 files
MEMORY.md is 26,691 bytes vs 24.4KB load limit (warning fires every session)
feedback memories are wired into Stop/UserPromptSubmit hooks

Claude Code commands cheat sheet YELLOW

What it is
A one-page list that explains handy Claude Code typing tricks, like the word "ultrathink" and the slash commands /context, /compact, /clear, and /rewind. It came from two YouTube videos Brandon watched.
Problem it solves
Brandon kept forgetting which built-in Claude Code command to use and when. The sheet matches each command to a rule he already follows, so he uses the right one without making up a new habit.
What Brandon wanted
He wanted a short map of the most useful Claude Code commands, lined up with his own rules (like "never default to Opus" and "notify every turn"), so the commands help his rules instead of adding clutter.
Connects to
/Users/drwu/Projects/_outputs/claude-code-adoption/commands-cheatsheet.md, /Users/drwu/Projects/_outputs/adoption-plans/claude-secret-codes-plan.md, /Users/drwu/.claude/settings.json, /Users/drwu/.claude/hooks/ultracode_default.py, /Users/drwu/.claude/_state/plans_enqueued.json, /Users/drwu/scripts/workflow_plans_bridge.py
Using it?
Partly. Half right. The commands the sheet talks about (ultrathink, /context, /compact, /clear, /rewind) are built into Claude Code, so Brandon can use them any time. The two on-disk things the sheet says are "already done" are real: the ntfy "Claude done" Stop hook and the ultracode default-on hook both live in settings.json. The sheet itself though is just a note. It is not a skill, it is not pinned anywhere, and it was never enqueued onto the plans board. So nothing new got wired in from this plan.
Why this color
The plan was written and is accurate, and the two hooks it points at (ntfy notify + ultracode default-on) really are live in settings.json. But the cheat-sheet itself was never wired in: it is not a skill, not referenced anywhere in ~/.claude, not pinned in Brandon-Vault, and never enqueued (plans_enqueued.json holds only the avoidant-narcissist plan). It is a reference note whose few action items were already covered by other work, so it is started-but-not-wired, not live-and-producing.
One tiny first step
Paste the four-line "two-line discipline" from the bottom of the sheet into ~/CLAUDE.md or pin the sheet in the Workflow Command Center, so the ultrathink/clear/compact habit is in front of Brandon every chat instead of buried in a one-off file.
Checked on disk
Read /Users/drwu/Projects/_outputs/claude-code-adoption/commands-cheatsheet.md (3889 bytes, created 2026-06-08, status: cheatsheet) — it is descriptive mapping, not an action list
/Users/drwu/.claude/_state/plans_enqueued.json contains ONLY the avoidant-narcissist-ai-video-pipeline plan; this cheat-sheet/secret-codes plan is NOT enqueued
settings.json line 80: Stop hook 'curl -H "Title: Claude done" -d "Done: $PWD" ntfy.sh/brandonmac' — the plan's Phase 2 notification claim is REAL and live
settings.json line 41: Notification hook to ntfy.sh/brandon-claude-9k3xq2 — confirms the attention-hook claim
settings.json line 115 + /Users/drwu/.claude/hooks/ultracode_default.py (605 bytes, exec) wired as UserPromptSubmit hook, default-on per Brandon's 2026-06-06 request — confirms the 'ultracode local default-on variant' claim
grep across /Users/drwu/.claude found NO reference to commands-cheatsheet, claude-secret-codes, or 'two-line discipline'; no skill dir matches *cheat*/*command*/*secret-code*
grep across Brandon-Vault found only one incidental session note, no dashboard pin or wired note
Source plan /Users/drwu/Projects/_outputs/adoption-plans/claude-secret-codes-plan.md exists (5494 bytes) with front-matter 'status: active' but never passed through ~/scripts/workflow_plans_bridge.py enqueue

The 7 Claude Code power commands YELLOW

What it is
A written plan to try 7 Claude Code shortcuts from a YouTube video. They are meant to save tokens and make daily work smoother (like ultrathink, /caveman, /insights, /btw, /clear, /loop, /schedule).
Problem it solves
Brandon spends money on tokens and wants his Claude Code days to run cleaner. These commands promise to cut waste and let him ask side questions without breaking a running job.
What Brandon wanted
To pick the few new commands that actually help, install them, measure the token savings, and write one house rule so future chats use them by default.
Connects to
/Users/drwu/.claude/hooks/ultracode_default.py, /Users/drwu/Projects/_outputs/claude-code-adoption/commands-cheatsheet.md, System Pulse (drwu-htmls.vercel.app/system-pulse), /session-audit skill, Brandon-Vault/40 Systems/Claude Code Prompts/, /Users/drwu/.claude/_state/plans_enqueued.json
Using it?
Partly. Half right. Phase 1 (the audit) really got done and written into the plan, and a real choice was made later: a cheatsheet says skip /caveman because it fights the voice and book gates. But that is the only piece that landed. The two NEW commands were never installed, nobody measured token savings, and the house rule never went into the vault. So the plan stalled after thinking, before doing.
Why this color
The plan was written and Phase 1 was audited, and a real keep/skip decision plus a cheatsheet exist. But Phase 2 and 3 never happened: no /caveman or /btw on disk, no token-delta measured, no house rule in the vault, and the plan was never enqueued. It is thought-through but not wired in, so it sits at the low end of yellow.
One tiny first step
Open the existing cheatsheet at /Users/drwu/Projects/_outputs/claude-code-adoption/commands-cheatsheet.md, copy its 5 keeper rules (ultrathink, /clear, /insights via /session-audit, /btw, skip caveman) into a new note at Brandon-Vault/40 Systems/Claude Code Prompts/claude-code-commands.md so the standing rule finally lives in the vault.
Checked on disk
Plan file exists: /Users/drwu/Projects/_outputs/claude-code-adoption/seven-claude-code-power-commands-plan.md with an inline 'Phase 1 RESULT executed 2026-06-06' section
plans_enqueued.json holds ONLY the avoidant-narcissist plan, not this one (so it was never enqueued)
ls /Users/drwu/.claude/skills/*caveman* and *btw* both return no matches; no commands/ files either
grep for caveman/btw in /Users/drwu/.claude/settings.json returns nothing
ultracode_default.py is real at /Users/drwu/.claude/hooks/ and matches the plan's ultrathink/ultracode claim
No token-delta measurement file exists anywhere in /Users/drwu/Projects/_outputs (the plan's stated 'measurable payoff')
Brandon-Vault/40 Systems/Claude Code Prompts/ contains only IDEA_MINER_PROMPT.md, so the Phase 3 house rule was never written there
commands-cheatsheet.md (created Jun 8, after the plan) folded in the keepers and recorded the decision to skip caveman because it 'fights the content work', but it lives in _outputs, not the vault

Obsidian core / PKM basics YELLOW

What it is
Videos teaching how to use Obsidian as one connected notebook: capture ideas fast, link notes, build hub pages instead of deep folders.
Problem it solves
Brandon's notes, plans, and dashboards live in many places. One linked vault means nothing gets lost and everything points to its related pieces.
What Brandon wanted
Obsidian as the project brain that ties every script, dashboard, prompt, and source together.
Connects to
~/Desktop/Brandon-Vault (numbered Workflow Hub folders), _Dashboards/Workflow Command Center.md, idea_miner.py daily job, 14 session notes in Sessions/
Using it?
Partly. The structure is built right (numbered folders, dashboards, a daily idea miner writing into it). The daily habit is thin: 14 session notes total and the dashboards are not opened most days.
Why this color
The vault is genuinely structured and machine-fed, and Brandon does not live in it yet. Built, not yet a habit.
One tiny first step
Pin Workflow Command Center and open it once each morning before starting work. One click, every day.
Checked on disk
Brandon-Vault has 00/10/20/30/40/90 numbered folders + _Dashboards
Sessions/ holds 14 notes
idea_miner launchd job writes 30 Distilled Ideas daily

Obsidian plugins (Excalidraw / Excalibrain) YELLOW

What it is
Videos about Obsidian add-ons: Excalidraw for drawing idea maps, Excalibrain for an automatic map of how notes connect, plus the handful of plugins worth keeping.
Problem it solves
Plain notes are hard to see as a whole. These plugins draw the picture so Brandon can SEE his system instead of remembering it.
What Brandon wanted
Visual maps of his skills, projects, and notes.
Connects to
Brandon-Vault/.obsidian/plugins/ (dataview, excalidraw, excalibrain, smart-connections, templater, quickadd, readwise, todoist-sync all installed), Skills Atlas.excalidraw
Using it?
Partly. Install is done: every plugin the videos recommend is already in the vault, and Dataview powers the Conductor Hub. The drawing tools (Excalidraw, Excalibrain) show little sign of regular use beyond the Skills Atlas.
Why this color
Everything is installed and one plugin (Dataview) earns its keep daily. The visual-map half of the method sits idle.
One tiny first step
Open Excalibrain once on the vault and look at the auto-map of one project. Zero setup, it is installed.
Checked on disk
.obsidian/plugins lists 13 community plugins including excalibrain + excalidraw
Conductor Hub.md is Dataview-driven
Skills Atlas.excalidraw exists

Claude Code + Obsidian second brain YELLOW

What it is
Videos showing Claude Code reading and writing an Obsidian vault, so chats leave notes behind and the vault becomes the memory between sessions.
Problem it solves
Chats vanish when they end. Wiring Claude to the vault means every chat deposits its work somewhere findable.
What Brandon wanted
Claude and Obsidian acting as one system, like the JARVIS videos he saved.
Connects to
chat-conductor skill (6 sub-skills, built 2026-05-20), UserPromptSubmit hooks, Brandon-Vault/Sessions/, Conductor Hub dashboard
Using it?
Partly. Brandon BUILT this method himself before watching more videos about it: chat-conductor routes chats into the vault with session notes, artifact logs, and Todoist push. Usage is the gap: most chats skip /chat-start, so the wiring sits unused (this very chat included).
Why this color
The system exists and works (14 session notes prove it ran). It needs the habit, not more building.
One tiny first step
Start the next chat with /chat-start <topic>. The machinery fires by itself after that.
Checked on disk
~/.claude/skills/chat-conductor with 6 sub-skills
Sessions/ has 14 notes
hooks inject conductor nudges every turn

n8n + local AI (Ollama / MCP) YELLOW

What it is
Videos about running AI on your own Mac (Ollama) and letting n8n robots call that free local AI instead of paid APIs.
Problem it solves
Paid API calls add up for bulk jobs. A local model costs nothing per run and works offline.
What Brandon wanted
A private, free AI layer for vault summaries and bulk text jobs.
Connects to
Ollama.app + 4 models (qwen2.5:3b, dolphin-mistral, llama2-uncensored, nomic-embed-text), n8n 'Claude -> Obsidian' + 'Starter: Summarize active note' workflows (built, switched off), local-gpt + smart-connections Obsidian plugins
Using it?
Partly. All the pieces are installed: Ollama with 4 models, n8n running, two local-AI workflows already built inside n8n, and two local-AI Obsidian plugins. Nothing is switched on, so the local models just take up 10GB of disk.
Why this color
Unusually far along (models downloaded, workflows written) with the final switch never flipped.
One tiny first step
In n8n, flip 'Starter: Summarize active note -> DeepSeek -> write to vault' to active and run it on one note.
Checked on disk
ollama list shows 4 models
n8n DB holds 'Claude -> Obsidian' and 'Starter: Summarize active note' at active=0
local-gpt plugin installed in vault

Official n8n MCP server RED

What it is
A plan to plug n8n's new official "MCP server" into Claude Code so Claude can build n8n robot-workflows by writing TypeScript code that gets checked before it touches the real n8n. It came from a YouTube video.
Problem it solves
Right now Brandon's n8n workflows are written by hand as raw JSON with no safety check, so it is easy to make a broken one. This would let Claude build and test them straight into the running n8n with no copy-paste.
What Brandon wanted
A safer, faster way to make n8n automations: tell Claude in plain English what he wants, and have Claude build a type-checked workflow right into his self-hosted n8n instead of guessing at JSON.
Connects to
~/Desktop/Brandon-Vault/40 Systems/n8n Workflows/ (idea_miner.py, n8n_daily_idea_mining_workflow.json, com.drwu.idea-miner.plist), ~/.n8n/ (self-hosted instance, com.drwu.n8n launchd job, database.sqlite), ~/.config/drwu/secrets.env (where the plan said to store the token), Claude Code MCP config (~/.claude.json / settings.json), ~/Projects/_outputs/claude-code-adoption/ (the 13-plan adoption folder)
Using it?
No. It is not being used at all, so there is nothing being done right or wrong yet. The plan is a clean, well-grounded write-up tied to the real video transcript, but none of its three phases ran. Phase 1 (turn on Instance-level MCP, add the server to Claude Code, see it in /mcp) was never done.
Why this color
Only the written plan exists. Hard checks all came back empty: ~/.claude.json has zero MCP servers (top-level mcpServers is []), settings.json has no n8n or mcp entries, secrets.env exists but holds no n8n token, ~/.n8n/.env has no MCP key, and plans_enqueued.json lists only the avoidant pipeline plan, not this one. The standing rule the plan said to write into the n8n Workflows folder was never written. The two skill hits (fleet-dashboard, system-pulse) only mention generic n8n/MCP status monitoring, not this plan.
One tiny first step
Open n8n's Settings, flip "Instance-level MCP" to Enabled, copy the URL and access token it gives you, and paste the token into ~/.config/drwu/secrets.env as N8N_MCP_TOKEN. That one step makes Phase 1 real.
Checked on disk
/Users/drwu/.claude/_state/plans_enqueued.json contains only avoidant-narcissist-ai-video-pipeline-plan.md, NOT the n8n plan
/Users/drwu/.claude.json: top-level mcpServers is empty ([]) and no per-project mcpServers found
grep of /Users/drwu/.claude/settings.json for 'n8n' and 'mcp' returned nothing
/Users/drwu/.config/drwu/secrets.env exists (695 bytes) but grep for n8n/N8N returned nothing
grep for 'mcp' in /Users/drwu/.n8n/.env returned nothing
n8n Workflows folder (~/Desktop/Brandon-Vault/40 Systems/n8n Workflows/) holds only the hand-authored JSON spec, idea_miner.py, and christian pipeline .ts; no MCP standing-rule note
grep across Brandon-Vault for 'official n8n / instance-level mcp / n8n mcp' hit only the avoidant plan and two session notes, no implemented rule
no /Users/drwu/.mcp.json file exists
skill hits fleet-dashboard/SKILL.md and system-pulse/index.html reference generic n8n/MCP status, not this plan's setup

Karpathy-style knowledge base RED

What it is
A plan to build a "second brain" wiki on top of Brandon's existing transcript piles. The idea, from a YouTube video about a Karpathy post, is three plain folders: raw notes Claude only reads, a wiki Claude writes, and a rulebook that says how the wiki is shaped.
Problem it solves
Right now Claude figures out the same things over and over from scratch every time you ask. Nothing gets saved and reused. This plan would let knowledge pile up so Claude does not start cold each time.
What Brandon wanted
Brandon watched the video and wrote up how to turn his big transcript folders (like the yt-claude-code one) into a cross-linked wiki, plus an auto step that updates the wiki when a new file is dropped in.
Connects to
/Users/drwu/New Desktop/yt-claude-code/transcripts/ (the pilot raw corpus, 35 transcripts, exists), /Users/drwu/.claude/projects/-Users-drwu/memory/MEMORY.md (the wiki would reference it), /Users/drwu/.claude/CLAUDE_LIBRARY_LOCATIONS.md (referenced as an existing index), iCloud sfg/inbox watcher (planned ingest landing spot), /Users/drwu/Desktop/Brandon-Vault (planned graph-view inspection)
Using it?
No. Not using it. The plan was written on 2026-06-06 and then nothing was built. There is no ~/Projects/_outputs/knowledge-base/ folder, no yt-claude-code wiki, no SCHEMA.md, no ingest skill, and no ingest hook in settings.json. The only piece that exists, the raw transcript folder, was already there before the plan. Brandon's own /plans board shows just 1 of 13 plans was ever enqueued, and this is not the one (plans_enqueued.json holds only the avoidant/narcissist video plan).
Why this color
It is only the written plan. I checked every place it said it would build something and found nothing: no knowledge-base folder, no SCHEMA.md, no wiki pages, no ingest skill, no hook wired into settings.json. The pilot raw folder exists but predates the plan, so it is not evidence of action. None of the three phases were started.
One tiny first step
Make the folder and the rulebook: run "mkdir -p ~/Projects/_outputs/knowledge-base/yt-claude-code" and have Claude write one SCHEMA.md (how a wiki page is shaped) plus one wiki page for a single topic cluster, with links back to the exact transcript filenames it cites. That alone finishes Phase 1.
Checked on disk
Plan file exists: /Users/drwu/Projects/_outputs/claude-code-adoption/karpathy-knowledge-base-plan.md (created 2026-06-06)
ls /Users/drwu/Projects/_outputs/knowledge-base/ -> NO knowledge-base FOLDER
find for SCHEMA.md under _outputs -> no result
find for any *knowledge*base* dir under ~/Projects and ~/.claude -> none
grep settings.json for knowledge.base/ingest/karpathy/wiki -> NO matches; hooks list has no ingest hook
ls ~/.claude/skills grep knowledge|ingest|karpathy|wiki -> NO matching skill folders
Karpathy grep hits in skills are unrelated (interview-me video, board-of-advisers IiZ5HRaeX4s.txt), not the knowledge-base videos yfeHoOkn2TI/hzQie4EucY0
plans_enqueued.json contains only one plan, the avoidant/narcissist video pipeline, not this one
Pilot raw layer exists: /Users/drwu/New Desktop/yt-claude-code/transcripts/ holds 35 transcripts, but it predates the plan and no wiki layer was built over it
Brandon-Vault grep 'knowledge base' hits are music run files and community-mirror JSON, not a Karpathy wiki

Claude Cowork courses RED

What it is
Video courses about Claude Cowork, the desktop mode where Claude works in shared folders alongside you like an employee.
Problem it solves
Some jobs fit a folder-watching coworker better than a chat. The courses teach when and how to use that mode.
What Brandon wanted
An AI employee setup better than the defaults, per the zero-to-AI-employee course he saved.
Connects to
setup-cowork skill (stock Anthropic, never run), 2026-05-21-firefox-tabs transcripts
Using it?
No. Not using it at all. The only traces on disk are the saved transcripts and the stock setup skill.
Why this color
Saved links only. No cowork output, no config, no run history anywhere on disk.
One tiny first step
Run the setup-cowork skill once on a small writing folder and see what the mode feels like.
Checked on disk
grep across Brandon-Vault hits only the transcript manifest
no cowork folder in ~/Projects/_outputs

AI agents full courses / agentic OS RED

What it is
Long video courses promising a complete education in AI agents and an 'agentic operating system' for your life.
Problem it solves
Courses promise a system for everything. Brandon already runs agents daily through Claude Code, so the courses mostly re-teach what he does.
What Brandon wanted
To check whether the courses held tricks his setup was missing.
Connects to
88 transcripts in 2026-05-21-firefox-tabs, existing Claude Code agent fleet (153 skills, workflows, hooks)
Using it?
No. The transcripts were captured and nothing was built FROM them. To be fair, Brandon's working system already covers most of what these courses teach.
Why this color
Saved transcripts only, zero artifacts derived from them. The links exist so the idea does not get lost, and nothing more happened.
One tiny first step
Have one chat skim the Nofar Gaspar agentic-OS transcript for the 3 ideas his system lacks, then queue ONE as a plan.
Checked on disk
transcripts saved (88 files in the firefox-tabs batch)
no output folder or plan cites these courses

Ableton automation RED

What it is
Videos teaching Ableton Live's automation lanes: making volume, filters, and effects move by themselves over time so loops sound alive.
Problem it solves
Brandon's generated loops and racks are static. Automation curves are the cheapest way to make a flat loop breathe.
What Brandon wanted
To fold automation moves into his MIDI-pack and drum-rack products.
Connects to
transcript DoRQo3aGaPY.txt in firefox-tabs, ableton-rack-builder + m4l-builder skills (adjacent tooling, built separately)
Using it?
No. Not using the method. The surrounding Ableton tooling is real and strong, and none of it does automation lanes.
Why this color
One saved transcript, no automation work product anywhere in the pack pipeline.
One tiny first step
Next time a loop pack renders, add ONE filter-sweep automation clip to the demo file using the video's first technique.
Checked on disk
transcript on disk matches 'automation'
no .als/.adg output uses automation lanes (rack builder writes static racks)
🛍️ Get the printable