Weekly best practice for second brains built in Anthropic’s Claude — Projects, memory, Obsidian & agents
Librairie.ai skunkworks · Masthead set in Fraunces · Body Newsreader · Labels Barlow Condensed
Lead · Tools
The line went onto the help page on 14 Oct, and MACCUS’s write-up (updated 18 Oct) was the first to spell out what it means. A second brain built as an old-style Project, a binder of chats plus a knowledge base and standing instructions, becomes a single conversation. A coordinator breaks each request into threads that keep running after you close the laptop, every thread starts with the project’s files, instructions and memory, and a Library collects what you add and what Claude produces.
Anthropic says ‘nothing is lost or deleted in the move’. Other chat and Cowork projects stay on the current version for now, and Team and Enterprise follow with no date. Neither the help page nor the docs says whether you can opt out or switch back, and the Claude Code docs, also updated 14 Oct, still describe a rollout limited to accounts with no existing projects.
What to do before your email arrives: tighten your project instructions so each thread starts with the right rules, prune the knowledge base so every thread isn’t loading stale files, and keep a copy of anything you can’t afford to lose outside Claude, for example in a Git repo or an Obsidian vault. Running several threads at once also uses your plan’s allowance faster.
Verified 19 Oct 2026 · items sourced from Anthropic, blogs, GitHub, YouTube & X
Anthropic’s ‘What are projects?’ help page now says it has started moving existing projects for some Pro and Max users to the redesigned version, where a project is one conversation that splits work into parallel cloud threads. Instructions, files, full chat history and scheduled tasks all carry over, and anyone who used a project in the last 30 days gets an email first.
6,142 likes · 529 reposts · 9,931 bookmarks
An 18 Oct post claims a developer ran 300 Claude Code agents in parallel on 100 company profiles, with a cheaper model checking every claim against its source, three passes until no errors were left, and the results written into an Obsidian vault. It drew 9.9k bookmarks and 1.7M views. The numbers are the poster’s claims; the useful idea, as @ggwplabs replied, is separating the writer from the checker and keeping the evidence in files outside the chat.
The 17 Oct release adds /crew: after every job each agent writes one line to a shared log in any folder or a vault, and session-start hooks hand each agent what the others did. Lessons agents leave for each other only become CLAUDE.md or AGENTS.md rules when you run keep. It also adds a source_quality label that /obsidian-health warns about when a note cites only speculation, plus German triggers.
AT Pulse’s 18 Oct brief, written up from a Reddit post and a nine-minute video, describes one orchestrator agent running a single setup prompt: it builds PARA-style folders, adds a ‘read the vault at startup’ pointer to each agent’s instruction file, and schedules a nightly logging job per agent, with one rules note in the vault governing how they all log. Small reach so far, but the same pattern as Crew above.
1,150 likes · 100 reposts · 2,517 bookmarks
24 likes · 0 reposts · 10 bookmarks
A quick tour of Claude Obsidian V2: a three-layer vault of persistent memory stored locally, an Obsidian knowledge graph linking it all, 15 skills that organise and verify notes, hooking it to an agent setup, and a 30-day roadmap. Heavy on course promotion, so treat it as a sketch rather than a manual.
Matt Connor’s 11 Oct guide runs the vault on a VPS inside tmux, so long refiling jobs survive a closed laptop, and syncs it with Syncthing. It covers CLAUDE.md rules (‘never rename or move a file’), Read+Edit deny rules for frozen folders, plan mode with useAutoModeDuringPlan off, and committing before every job so git diff --stat shows exactly what Claude touched. Hosting-vendor guide.
204 likes · 17 reposts · 329 bookmarks
166 likes · 49 reposts · 202 bookmarks
Fadel Kaadan’s guide (updated 10 Oct): run Claude Code inside your notes folder, write a CLAUDE.md that says how the vault is organised, how Claude should write and what it must report back, put the vault in Git and review diffs before committing. Vendor guide — it ends with a pitch for Knowtarium’s MCP workspace.
On 9 Oct the Claude Code team let every Pro and Max user on the waitlist into the redesigned Projects. That post got 15.6k likes and 15.1k bookmarks. The Projects guide, revised 10 Oct, adds the practical limits that matter for a second brain: uploaded folders are copies (first 100 files, 200 MB) that don’t sync, and a project with several repos on Anthropic’s hosted environment applies no repository’s permission rules.
A 9 Oct practice guide built on Anthropic’s ‘new rules of context engineering’ post. Anthropic cut over 80% of Claude Code’s system prompt for Claude 5 models, so your CLAUDE.md should hold only what Claude can’t work out itself. Move procedures into skills, let auto memory record corrections instead of the old # hotkey, and run /doctor prompt-audit (v2.1.283+) to flag stale or contradictory instruction files.
This week’s releases (5–9 Oct): path-scoped rules and nested CLAUDE.md files now load when Claude views a file with a single cat, head, tail, sed -n or grep, not only the Read tool (2.1.293). A subfolder’s AGENTS.md attaches when a file under it is @-mentioned (2.1.290), symlinked CLAUDE.md files outside the working directory respect read blocks, and synced claude.ai skills check for changes every ~40 min when idle.
15613 likes · 810 reposts · 15110 bookmarks
82 likes · 6 reposts · 130 bookmarks
Anthropic’s own quick start, posted as every Pro and Max waitlist user was let in: create a project, send work, and let threads share memory.
30 likes · 2 reposts · 43 bookmarks
James Gray (9 Oct) maps the three releases that reshaped where a Claude second brain lives: unified memory (25 Aug), the chat/Cowork merge (16 Sep) and cloud execution for every new Pro/Max task (6 Oct). His advice: keep a changing knowledge graph in a synced GitHub or Drive source, not a one-time cloud upload.
Keep your context out of any one vendor’s cloud: push it to a GitHub repo as a decoupled context layer, connect that repo to Claude Projects (and the catch with Library and Google Drive), then test read/write from ChatGPT too. Ends on scheduled tasks, teams and memory.
Explainer (8 Oct): every cloud thread in a Claude Code project reads the MEMORY.md index at start and can add to it, so a correction made once reaches later threads. Project memory is per user and per project, and is separate from local Claude Code auto memory; threads run locally via Remote Control get instructions but not the memory files.
Haiku 5.5 (7 Oct) is aimed at high-volume work like summaries, compactions, classification and subagents: $0.10/$0.50 per million tokens for prompts up to 100K, 1M context and 128K output, about 75% cheaper than Haiku 4.5. That makes it a fit for bulk ingest and tagging passes over a vault. Anthropic also halved Sonnet 5.5 cache-read prices and added monthly API credits for Max ($100/$200) and Team (up to $500).
106 likes · 14 reposts · 124 bookmarks
64 likes · 8 reposts · 17 bookmarks
217 likes · 26 reposts · 349 bookmarks
Comparison guide: Claude Code in the vault folder needs no plugin but has no guard against stale overwrites or broken links; the official Obsidian CLI (1.12.7+) gives link-aware renames but needs a desktop session; the Local REST API plugin’s built-in MCP server (v5.4.0, 6 Oct) is another route. Written by a vendor selling its own CLI.
Four steps to turn 700,000 words of one expert’s lectures, blogs and posts into a Claude Code agent that teaches the way he does. The same method works for any expert corpus you keep in your second brain.
153 likes · 28 reposts · 64 bookmarks
Nolen Jonker (4 Oct) on letting Claude Code do the linking Obsidian users never keep up with: it surfaces connections between notes written months apart, finds recurring themes across the whole vault, ranks notes by what will matter later, and quizzes you from your own notes.
Official Projects guide (revised 3 Oct): every cloud thread reads the project’s MEMORY.md index at start and opens topic files on demand; project instructions reach every thread (up to 16,000 characters); repo CLAUDE.md files still load from each clone. Project memory is separate from local Claude Code auto memory. Pro/Max beta; 200 new threads/day cap.
234 likes · 22 reposts · 251 bookmarks
v2.1.284–2.1.288 (28 Sep–2 Oct): MEMORY.md and recalled notes now neutralise invisible characters and fake Claude Code markup; Auto-memory can’t be switched on from background sessions; /memory toggles with arrow keys; CLAUDE.md no longer attaches twice after resume/compaction; path-scoped rules and nested CLAUDE.md now load on Write/Edit, not just Read.
Claude Code v2.1.287 (1 Oct) adds mods: small TypeScript functions, shipped inside plugins, that hook events to rewrite prompts or tool calls, add UI or replace built-ins (/diff is now a mod). Claude can write and hot-reload one for you. Mods run with Claude Code’s full access to your machine — install only from sources you trust.
Plain-markdown vault as the whole productivity stack: regenerated this-week.md and stalled.md dashboards, YAML Bases Claude can write directly, and a CLAUDE.md rule to save research answers in /Cache/ and check there before going back to sources. Collaboration, design and reminders still need other apps.
Guide updated 1 Oct, built on Karpathy’s LLM-wiki layout: inbox/, read-only raw/, notes/, plus index.md and an append-only log.md. Deny rules like Edit(/raw/**) in .claude/settings.json, @AGENTS.md import for Codex, Obsidian CLI move/rename to keep links intact, and an /inbox skill. Written by a vendor that sells a hosted agent wiki.
3 likes · 0 reposts · 1 bookmarks
RAG instead of a markdown pile: Claude Cowork writes its own upload scripts, loads 15 NASA videos into a searchable store, then answers ‘which video said…?’ with a link to the source. Sponsored by Progress Agentic RAG.
49 likes · 10 reposts · 41 bookmarks
Anthropic, 28 September 2026: Sonnet 5.5 is a faster, lower-cost complement to Opus 5.5. It runs 30%+ faster than Sonnet 5 and costs up to 30% less for most work, at the same $2 / $10 per million input / output token prices, because it typically uses fewer tokens. Strongest, Anthropic says, at well-scoped everyday tasks, bug fixes, and polished documents, slides and spreadsheets. Model id claude-sonnet-5-5.
52 likes · 11 reposts · 24 bookmarks
6 likes · 1 reposts · 17 bookmarks
1251 likes · 177 reposts · 2826 bookmarks
Classic Projects path: knowledge uploads, custom instructions, RAG expansion up to ~10× on paid plans, Team/Enterprise sharing. Free users still capped at five classic projects while the parallel-thread beta rolls out via Claude Code.
96 likes · 18 reposts · 103 bookmarks
1482 likes · 213 reposts · 3067 bookmarks
45 likes · 2 reposts · 64 bookmarks
Official guide (updated mid-Sep): memory saves as topics while you chat; each Project gets its own memory space and summary; paid users can RAG-search past chats; legacy memory export window ran through 9 Sep 2026. Reset memory wipes project memories too.
Updated 21 Sep guide: durable notes stay authoritative; CLAUDE.md sets boundaries; Generated/ quarantines drafts; daily plan + weekly review loops with source citations. Clarifies that a vault memory.md is not Claude Code auto memory — use /memory to inspect the real store.
Hands-on with the redesigned Projects: one coordinator spinning up threads that share memory, artifacts and routines; three business projects built in 48 hours; seven things to know, including tracking token use per thread.
302 likes · 30 reposts · 295 bookmarks
Official docs: human-written CLAUDE.md / AGENTS.md for rules, plus auto memory Claude writes itself. MEMORY.md loads the first 200 lines or 25KB each session; keep CLAUDE.md under ~200 lines; path-scoped .claude/rules/ load on demand.
Official blog: Claude Code Projects now run a coordinator plus parallel cloud threads that keep working after you close the laptop. Threads share project memory and a library of files/artifacts; Pro/Max beta expands this week, then chat, Cowork, Team and Enterprise.
Stevie Bonifield reports the 17 Sep relaunch: coordinator + parallel threads with shared memory, goals and artifact library; each thread is a cloud Claude Code session on its own branch; local tooling promised ‘very soon’.
Architecture piece (updated 28 Aug): PARA vault as OS, seven-section CLAUDE.md, session hooks, and a ~500-word hot.md snapshot the agent reads first every session. Five-step interview to generate a vault-specific operating manual; MCP layers only after the manual loop works.
Argues Claude beats ChatGPT for Obsidian PKM because Claude Code auto-loads CLAUDE.md, Projects persist file context, and Claude Skills (incl. Steph Ango’s Obsidian skills) compose reusable workflows. Warns against over-stuffing vaults and skipping CLAUDE.md.
82 likes · 12 reposts · 123 bookmarks
Sam McKay’s production walkthrough: instructions vs knowledge, curation over volume, conversation isolation (chats don’t inherit prior chat memory), and the lack of live sync — re-upload when source docs change.
Deep dive: managed/user/project/local CLAUDE.md scopes; Auto Memory + MEMORY.md 200-line/25KB cap; Auto Dream consolidation per store; native workarounds (user rules, symlinked .claude/rules/) before reaching for external backends.
PARA-style vault (Inbox/Projects/Areas/Resources/Archive/AI/Templates) plus a CLAUDE.md that defines session start/end protocols and where the agent may write. Emphasises closing the loop by writing session summaries back into the vault.
Wyndo maps Tiago Forte’s PARA into an Obsidian vault and wires Claude Code as a file-native copilot. Memory layer: CLAUDE.md, VAULT-INDEX.md, SessionStart hook. Commands: /plan-week, /daily-prep, /process-inbox, /end-day, /review-week.
Articles & official docs
Ranked by engagement
YouTube embeds + links
Docs, GitHub, community
An 18 Oct post claims a developer ran 300 Claude Code agents in parallel on 100 company profiles, with a cheaper model checking every claim…
AT Pulse’s 18 Oct brief, written up from a Reddit post and a nine-minute video, describes one orchestrator agent running a single setup prom…
The 17 Oct release adds /crew: after every job each agent writes one line to a shared log in any folder or a vault, and session-start hooks…
Four releases from 14 to 16 Oct. 2.1.302 adds an ‘Idle compaction’ row to /config so long-lived vault sessions are not summarised while you…
A 16 Oct list (814 likes, 1.6k bookmarks) argues that a pile of notes isn’t memory and names seven open-source repos, graphify among them, w…
A 14 Oct thread telling people to make a folder called SECOND BRAIN, launch Claude Code inside it and install the Google Workspace CLI so it…
An 18 Oct post claims a developer ran 300 Claude Code agents in parallel on 100 company profiles, with a cheaper model checking every claim against its source, three passes until no errors were left, and the results written into an Obsidian vault. It drew 9.9k bookmarks and 1.7M views. The numbers are the poster’s claims; the useful idea, as @ggwplabs replied, is separating the writer from the checker and keeping the evidence in files outside the chat.
AT Pulse’s 18 Oct brief, written up from a Reddit post and a nine-minute video, describes one orchestrator agent running a single setup prompt: it builds PARA-style folders, adds a ‘read the vault at startup’ pointer to each agent’s instruction file, and schedules a nightly logging job per agent, with one rules note in the vault governing how they all log. Small reach so far, but the same pattern as Crew above.
A 14 Oct thread telling people to make a folder called SECOND BRAIN, launch Claude Code inside it and install the Google Workspace CLI so it can reach Gmail, Calendar and Drive drew 2.9k likes and 6.1k bookmarks, the most-saved second-brain post this week. Within three days at least five near-identical threads followed.
Matt Connor’s 11 Oct guide runs the vault on a VPS inside tmux, so long refiling jobs survive a closed laptop, and syncs it with Syncthing. It covers CLAUDE.md rules (‘never rename or move a file’), Read+Edit deny rules for frozen folders, plan mode with useAutoModeDuringPlan off, and committing before every job so git diff --stat shows exactly what Claude touched. Hosting-vendor guide.
Fadel Kaadan’s guide (updated 10 Oct): run Claude Code inside your notes folder, write a CLAUDE.md that says how the vault is organised, how Claude should write and what it must report back, put the vault in Git and review diffs before committing. Vendor guide — it ends with a pitch for Knowtarium’s MCP workspace.
On 9 Oct the Claude Code team let every Pro and Max user on the waitlist into the redesigned Projects. That post got 15.6k likes and 15.1k bookmarks. The Projects guide, revised 10 Oct, adds the practical limits that matter for a second brain: uploaded folders are copies (first 100 files, 200 MB) that don’t sync, and a project with several repos on Anthropic’s hosted environment applies no repository’s permission rules.
A 9 Oct practice guide built on Anthropic’s ‘new rules of context engineering’ post. Anthropic cut over 80% of Claude Code’s system prompt for Claude 5 models, so your CLAUDE.md should hold only what Claude can’t work out itself. Move procedures into skills, let auto memory record corrections instead of the old # hotkey, and run /doctor prompt-audit (v2.1.283+) to flag stale or contradictory instruction files.
James Gray (9 Oct) maps the three releases that reshaped where a Claude second brain lives: unified memory (25 Aug), the chat/Cowork merge (16 Sep) and cloud execution for every new Pro/Max task (6 Oct). His advice: keep a changing knowledge graph in a synced GitHub or Drive source, not a one-time cloud upload.
Nolen Jonker (4 Oct) on letting Claude Code do the linking Obsidian users never keep up with: it surfaces connections between notes written months apart, finds recurring themes across the whole vault, ranks notes by what will matter later, and quizzes you from your own notes.
Official Projects guide (revised 3 Oct): every cloud thread reads the project’s MEMORY.md index at start and opens topic files on demand; project instructions reach every thread (up to 16,000 characters); repo CLAUDE.md files still load from each clone. Project memory is separate from local Claude Code auto memory. Pro/Max beta; 200 new threads/day cap.
Plain-markdown vault as the whole productivity stack: regenerated this-week.md and stalled.md dashboards, YAML Bases Claude can write directly, and a CLAUDE.md rule to save research answers in /Cache/ and check there before going back to sources. Collaboration, design and reminders still need other apps.
Guide updated 1 Oct, built on Karpathy’s LLM-wiki layout: inbox/, read-only raw/, notes/, plus index.md and an append-only log.md. Deny rules like Edit(/raw/**) in .claude/settings.json, @AGENTS.md import for Codex, Obsidian CLI move/rename to keep links intact, and an /inbox skill. Written by a vendor that sells a hosted agent wiki.
Anthropic, 28 September 2026: Sonnet 5.5 is a faster, lower-cost complement to Opus 5.5. It runs 30%+ faster than Sonnet 5 and costs up to 30% less for most work, at the same $2 / $10 per million input / output token prices, because it typically uses fewer tokens. Strongest, Anthropic says, at well-scoped everyday tasks, bug fixes, and polished documents, slides and spreadsheets. Model id claude-sonnet-5-5.
Official guide (updated mid-Sep): memory saves as topics while you chat; each Project gets its own memory space and summary; paid users can RAG-search past chats; legacy memory export window ran through 9 Sep 2026. Reset memory wipes project memories too.
Updated 21 Sep guide: durable notes stay authoritative; CLAUDE.md sets boundaries; Generated/ quarantines drafts; daily plan + weekly review loops with source citations. Clarifies that a vault memory.md is not Claude Code auto memory — use /memory to inspect the real store.
Official docs: human-written CLAUDE.md / AGENTS.md for rules, plus auto memory Claude writes itself. MEMORY.md loads the first 200 lines or 25KB each session; keep CLAUDE.md under ~200 lines; path-scoped .claude/rules/ load on demand.
Official blog: Claude Code Projects now run a coordinator plus parallel cloud threads that keep working after you close the laptop. Threads share project memory and a library of files/artifacts; Pro/Max beta expands this week, then chat, Cowork, Team and Enterprise.
Architecture piece (updated 28 Aug): PARA vault as OS, seven-section CLAUDE.md, session hooks, and a ~500-word hot.md snapshot the agent reads first every session. Five-step interview to generate a vault-specific operating manual; MCP layers only after the manual loop works.
Argues Claude beats ChatGPT for Obsidian PKM because Claude Code auto-loads CLAUDE.md, Projects persist file context, and Claude Skills (incl. Steph Ango’s Obsidian skills) compose reusable workflows. Warns against over-stuffing vaults and skipping CLAUDE.md.
Sam McKay’s production walkthrough: instructions vs knowledge, curation over volume, conversation isolation (chats don’t inherit prior chat memory), and the lack of live sync — re-upload when source docs change.
Deep dive: managed/user/project/local CLAUDE.md scopes; Auto Memory + MEMORY.md 200-line/25KB cap; Auto Dream consolidation per store; native workarounds (user rules, symlinked .claude/rules/) before reaching for external backends.
Wyndo maps Tiago Forte’s PARA into an Obsidian vault and wires Claude Code as a file-native copilot. Memory layer: CLAUDE.md, VAULT-INDEX.md, SessionStart hook. Commands: /plan-week, /daily-prep, /process-inbox, /end-day, /review-week.
The 17 Oct release adds /crew: after every job each agent writes one line to a shared log in any folder or a vault, and session-start hooks hand each agent what the others did. Lessons agents leave for each other only become CLAUDE.md or AGENTS.md rules when you run keep. It also adds a source_quality label that /obsidian-health warns about when a note cites only speculation, plus German triggers.
Four releases from 14 to 16 Oct. 2.1.302 adds an ‘Idle compaction’ row to /config so long-lived vault sessions are not summarised while you are away. The desktop app and VS Code now always run the main conversation’s subagents in the background, as the terminal does. 2.1.301 fixes skills no longer being listed to the model after /rewind, and switching to a model with a smaller context window now has the old model summarise the conversation first.
A 16 Oct list (814 likes, 1.6k bookmarks) argues that a pile of notes isn’t memory and names seven open-source repos, graphify among them, where Claude reads a source once and links it to everything already in the vault. It follows Andrej Karpathy’s description of dropping papers, articles and repos into one folder and letting the LLM write and maintain a wiki on top of it.
Anthropic’s ‘What are projects?’ help page now says it has started moving existing projects for some Pro and Max users to the redesigned version, where a project is one conversation that splits work into parallel cloud threads. Instructions, files, full chat history and scheduled tasks all carry over, and anyone who used a project in the last 30 days gets an email first.
On 14 Oct the Claude Code team said the desktop app now starts sessions much faster, makes it easier to split work across several, and shows what each one is doing (20.5k likes). On 16 Oct it added that Claude Code now gzips the conversation it sends each turn and cloud sessions start as soon as you type. Useful for running a filing session, an inbox triage and a weekly review in parallel over one vault.
This week’s releases (5–9 Oct): path-scoped rules and nested CLAUDE.md files now load when Claude views a file with a single cat, head, tail, sed -n or grep, not only the Read tool (2.1.293). A subfolder’s AGENTS.md attaches when a file under it is @-mentioned (2.1.290), symlinked CLAUDE.md files outside the working directory respect read blocks, and synced claude.ai skills check for changes every ~40 min when idle.
Explainer (8 Oct): every cloud thread in a Claude Code project reads the MEMORY.md index at start and can add to it, so a correction made once reaches later threads. Project memory is per user and per project, and is separate from local Claude Code auto memory; threads run locally via Remote Control get instructions but not the memory files.
Haiku 5.5 (7 Oct) is aimed at high-volume work like summaries, compactions, classification and subagents: $0.10/$0.50 per million tokens for prompts up to 100K, 1M context and 128K output, about 75% cheaper than Haiku 4.5. That makes it a fit for bulk ingest and tagging passes over a vault. Anthropic also halved Sonnet 5.5 cache-read prices and added monthly API credits for Max ($100/$200) and Team (up to $500).
Comparison guide: Claude Code in the vault folder needs no plugin but has no guard against stale overwrites or broken links; the official Obsidian CLI (1.12.7+) gives link-aware renames but needs a desktop session; the Local REST API plugin’s built-in MCP server (v5.4.0, 6 Oct) is another route. Written by a vendor selling its own CLI.
v2.1.284–2.1.288 (28 Sep–2 Oct): MEMORY.md and recalled notes now neutralise invisible characters and fake Claude Code markup; Auto-memory can’t be switched on from background sessions; /memory toggles with arrow keys; CLAUDE.md no longer attaches twice after resume/compaction; path-scoped rules and nested CLAUDE.md now load on Write/Edit, not just Read.
Claude Code v2.1.287 (1 Oct) adds mods: small TypeScript functions, shipped inside plugins, that hook events to rewrite prompts or tool calls, add UI or replace built-ins (/diff is now a mod). Claude can write and hot-reload one for you. Mods run with Claude Code’s full access to your machine — install only from sources you trust.
Classic Projects path: knowledge uploads, custom instructions, RAG expansion up to ~10× on paid plans, Team/Enterprise sharing. Free users still capped at five classic projects while the parallel-thread beta rolls out via Claude Code.
Stevie Bonifield reports the 17 Sep relaunch: coordinator + parallel threads with shared memory, goals and artifact library; each thread is a cloud Claude Code session on its own branch; local tooling promised ‘very soon’.
PARA-style vault (Inbox/Projects/Areas/Resources/Archive/AI/Templates) plus a CLAUDE.md that defines session start/end protocols and where the agent may write. Emphasises closing the loop by writing session summaries back into the vault.
GitHub — curated for Claude second-brain builders.
GitHub — curated for Claude second-brain builders.
GitHub — curated for Claude second-brain builders.
GitHub — curated for Claude second-brain builders.
GitHub — curated for Claude second-brain builders.
‘We just let in every Pro and Max user from the Claude Code Projects waitlist!’ — with a 4-minute walkthrough for anyone new to Projects.
OPEN ON X →‘We’ve been working hard on making the Claude Code desktop app great for running lots of sessions side by side’: quicker starts, easier splitting, and you can always see what each session is doing.
OPEN ON X →‘300 Claude Code agents in parallel… hooked them to a verification layer so none of them could bullshit him’, with every result dumped into an Obsidian vault.
OPEN ON X →‘Create a folder named SECOND BRAIN on your PC. Launch Claude Code inside this folder and install GWS CLI’, the one-folder recipe that gives Claude your Gmail, Calendar and Drive.
OPEN ON X →Viral ‘free gold’ pack: a second-brain article hit 8M views, so the author published the full setup — repo, 65-page guide, tools and learning path — for free.
OPEN ON X →Lists 20 GitHub projects for a system that captures sources, connects ideas, remembers context and helps agents act — ‘your second brain should get sharper after every session.’
OPEN ON X →‘Claude can now read your X bookmarks, and turn the 2,000 posts you never reread into a second brain’: connect your X account and run seven prompts.
OPEN ON X →‘Stop storing notes. Start compiling them’: seven repos where Claude reads a source once and links it to everything you already have.
OPEN ON X →CLAUDE + OBSIDIAN + LOOP ENGINEERING: four-step loop where Claude opens the vault (not a chat window) — framed as close to Karpathy’s second-brain workflow.
OPEN ON X →Karpathy’s llm-wiki as one prompt for Opus 5.5: Claude builds a Zettelkasten-style second brain of linked notes, browsable in Obsidian.
OPEN ON X →‘The masterclass for a Karpathy-style second brain in Obsidian, Zettelkasten included’: the 8M-view article’s author packed repo, guide, tools and learning path into one free repo.
OPEN ON X →‘How to build a second brain with Opus 5.5 in under an hour’: download Obsidian, create 8 folders, open Claude Code inside the vault, write one CLAUDE.md, type /process.
OPEN ON X →‘Claude has a feature called Projects. 99% of users ignore it’: how to set up five Projects as a second brain for your whole business.
OPEN ON X →‘Steal my second brain CLAUDE.md’: one idea per note, keep my words, never delete — archive instead, cite every note you use.
OPEN ON X →‘Second Brain OS’: everything written and collected on AI agents laid out as a clickable map of linked pages, terms and tools.
OPEN ON X →Vendor claim (Leviathan 0.2.0): after 300 sessions an agent’s MEMORY.md costs 33K tokens per prompt and serves an outdated fact alongside 3 in 4 answers — a case for pruning.
OPEN ON X →Follow-up loop post: Claude Opus opens the Obsidian vault on a four-step read → write → link → remember cycle — almost none of the vault typed by hand.
OPEN ON X →Earlier viral riff on a free guide for building a second brain with Claude and Obsidian that ‘never forgets’ — still a useful community proof point.
OPEN ON X →NEW TUTORIAL: Claude Code Projects Clearly Explained in 39 Minutes — Anthropic’s new Projects paradigm spins up multiple cloud agents from one brief.
OPEN ON X →‘One weekend to master Claude’: a short learning path through Anthropic’s own courses — Claude 101, Claude Code 101, Claude Code in Action — instead of bookmarking 50 guides.
OPEN ON X →‘40 lines of text turn Obsidian into an Opus 5.5 second brain’: no plugins, no code — just a CLAUDE.md in the vault root.
OPEN ON X →Amplifies the free second-brain repo drop: 10 sections / 65 pages covering concept through troubleshooting after an 8M-view article.
OPEN ON X →20 GitHub projects for a second brain that gets smarter every session, grouped by capture, organise and recall.
OPEN ON X →‘Stop building your AI second brain inside an app’: notes/, people/, projects/, MEMORY.md, LEARNINGS.md and decisions.md, with CLAUDE.md on top as the map. One file, one topic.
OPEN ON X →Frames Claude Projects as Anthropic baking in ‘The Orchestrator Rule’: tell a coordinator agent what you want; it spins up and manages worker agents.
OPEN ON X →‘Your second brain doesn’t need Obsidian, Notion or a database’: notes/, people/, projects/, one file per topic, and one CLAUDE.md that says where everything lives.
OPEN ON X →Hands-on test of eugeniughelbur/obsidian-second-brain — persistent memory for Claude Code and six other CLI agents as plain markdown in an Obsidian vault: ‘nothing broke, ran first time.’
OPEN ON X →A quick tour of Claude Obsidian V2: a three-layer vault of persistent memory stored locally, an Obsidian knowledge graph linking it all, 15 skills that organise and verify notes, hooking it to an agent setup, and a 30-day roadmap. Heavy on course promotion, so treat it as a sketch rather than a manual.
A walk through the rebuilt Projects from the ground up: the coordinator agent that splits a prompt into threads and returns one plan, how to set a project up so the coordinator works toward your goal, and putting recurring work on a schedule.
A five-minute, no-terminal setup: install Obsidian, connect it to Claude, test the connection, then move your business, clients, goals and decisions into the vault so Claude can find them in every new chat.
When a vault is about 7 million tokens, more than Claude Code can read at once, Zhutov hands it a NotebookLM library of sources instead, with a NotebookLM skill as the bridge.
Keep your context out of any one vendor’s cloud: push it to a GitHub repo as a decoupled context layer, connect that repo to Claude Projects (and the catch with Library and Google Drive), then test read/write from ChatGPT too. Ends on scheduled tasks, teams and memory.
Four steps to turn 700,000 words of one expert’s lectures, blogs and posts into a Claude Code agent that teaches the way he does. The same method works for any expert corpus you keep in your second brain.
Anthropic’s own quick start, posted as every Pro and Max waitlist user was let in: create a project, send work, and let threads share memory.
RAG instead of a markdown pile: Claude Cowork writes its own upload scripts, loads 15 NASA videos into a searchable store, then answers ‘which video said…?’ with a link to the source. Sponsored by Progress Agentic RAG.
Hands-on with the redesigned Projects: one coordinator spinning up threads that share memory, artifacts and routines; three business projects built in 48 hours; seven things to know, including tracking token use per thread.
Vault brief in ~15 seconds each session, Obsidian CLI skill so Claude can create/move/search notes, and a live content-creator workflow from raw idea to researched note.
Free setup: open Claude Code inside the vault, interview yourself to personalise CLAUDE.md, and keep the vault folder scoped so the agent can’t roam the whole machine.
Live walkthrough of Forte’s AI Second Brain: master prompt, PARA context bundles, and using Claude Code to refresh the master prompt from notes.
Builds a weekly Monday brief skill, runs parallel subagents over the vault graph, and wires Gmail/Calendar/Drive so the second brain can act — not just store.
Self-improving knowledge base pattern (raw / wiki / outputs + CLAUDE.md) using Claude Cowork / Code as librarian.
39-minute tutorial on the new Claude Code Projects parallel-agent paradigm (linked from the highly engaged X post).
Items were gathered via WebSearch / WebFetch from Anthropic, Claude Code docs and changelog, independent blogs, GitHub, YouTube (IDs verified via oEmbed) and the X API (search_posts_all). Engagement metrics on X posts are public_metrics at collection time. Stats cards only use numbers stated in those sources. Layout mirrors Librairie’s The Treatment hub; masthead face is Fraunces (not the classic editorial serif used there).
Machine-readable feed: feed.json (bundled with this page).