Founder OS - Playbook
The Playbook

Your Second Brain

The why behind Founder OS. Set up now or read first. 15 minutes here saves a week of trying things that don't work.

Running a business or working toward founding one - this OS reads for both. There's a section for that.

Start setting up →
6 operating files
58 skills
30 min to set up
Any AI model works

Start setting up your brain in less than 30 seconds.

30 seconds to start · 30 minutes to set up · 30 days to compound

< 30 seconds
Click download on the FounderOS repo Zip file downloaded Extract the zip file
Right-click Extract All...

Download and unzip.

< 1 minute
Choose the FounderOS folder in Claude Code

Open the folder in Claude Code.

You're live
First prompt: Help me create my second brain

Say: Help me create my second brain.

The problem

You're running alone. Or close to it.

Every decision goes through you. Every client question. Every invoice. Every hire. Every piece of work that leaves the door.

You built it this way - not on purpose, but because you were the best person for every job when the business was small. And now the business grew around you. And now you can't leave.

Not for a week. Barely for a day.

Client questions Hiring decisions Deal approvals Quality checks Invoice sign-offs You bottleneck Nothing moves without you.
Now picture the after.

Things keep moving because the brain holds what only you held. One layer. You feed it three lines a day. It does the rest. The week starts informed, not chasing. Friday dinner doesn't get interrupted by a thread you forgot to follow up on.

This isn't a productivity problem. It's a context problem. The business runs because you hold everything in your head. Your standards. Your relationships. Your instincts about what to pursue. Nobody else has that. And because nobody else has it, nothing runs without you.

The tools haven't helped. You've tried Notion. Trello. Monday. Asana. The problem was never the tool. The problem is that every tool required you to maintain it. Another inbox. Another place to check. Another thing that goes stale when you get busy.

So you stop using it. And the cycle repeats.

Here's what's actually happening. Every time you open a new tool, you're creating a second brain. And a second brain competes with the first one. You spend more energy tracking two systems than either system is saving you.

What this is

You know your business, your team, and your clients better than anyone. You built all of this. Founder OS doesn't replace what you hold. It gives you a place to put it down so the system can carry it with you.

Founder OS is an operating layer that lives wherever you talk to AI. It's a folder on your machine that any AI tool can read.

We recommend Claude Code - the same Claude you already know, in a tab that can read your whole folder automatically. You point it at the folder. It sees everything. You don't paste anything.

Think of Claude Code less as a developer tool and more as the version of Claude that already knows you. Same model. Same chat. More context. The Claude desktop app, the Claude web app, Codex, and Cursor all work too - they just need you to bring the files into the conversation. The OS is the files. The tool is your choice.

The difference between a chat app and Claude Code: the chat is a fresh conversation. Claude Code is a fresh conversation that already knows you.
Built for non-technical founders.

If you've never opened a terminal, you don't need to. The setup wizard talks to you in plain English. You answer. It builds. Everything is a plain text page you can read and edit. There's nothing here that can break in a way two clicks can't undo.

Six files. That's the whole brain.

Who you are  ·  What you're focused on right now  ·  Open decisions  ·  Your active clients  ·  What's on for today  ·  What this week needs

Identity

Priorities

Decisions

Clients

Daily anchors

Weekly commit

Six plain text pages. Human-readable, AI-readable, yours forever.

From those six pages, 58 skills come alive. Skills that prep you for meetings, draft emails in your voice, log what happened, flag what you're avoiding, help you decide when you're stuck. You don't command them. You talk. They fire.

And once a week, the whole thing resets. You run the retro. You roll what's done. You park what's not. And you start Monday knowing exactly what matters.

The OS is yours, not automation. Every decision you log, every flag you track, every rant you capture - a human ran through it. The AI does the formatting. The judgment is yours.

Why not a dedicated AI workflow tool?

Those are good tools. The reason to start here is that it's easier - you can rant your way out of a problem instead of writing a config. Once you know how you actually work with an AI, you can move to anything. The files come with you.

If you're not running yet

The OS reads the same way for someone trying to get to founding as for someone already running. The difference is which skills you lean on first.

Before there are clients, the brain layer holds your evolving thesis instead of accounts. The cadence keeps you honest about whether you're actually testing the idea or just talking about it. And the same skills founders use to ship help you frame the problem, validate it, and decide whether to commit.

You're not running yet. But the brain you build now is the same brain you'll need on day one of running. Skip ahead in this playbook to the seven-day setup. The files you fill won't go to waste.

Four layers

Hold it in your head as three layers stacked on top of each other, with one rhythm around them.

The Brain is what your business knows about itself. Who matters, what's open, what was decided. Six text files. Nothing fancy.

The Skills are abilities the brain has. Write an email. Prep a meeting. Draft a proposal. Each skill reads the brain before it writes anything.

The Hands are the tools the skills reach for. Your calendar, your inbox, your notes, your transcriber. The OS doesn't replace them. It connects them.

Around all three, a heartbeat. A daily anchor. A weekly retro. Without it, the brain goes stale and the skills stop firing clean.

The Heartbeat daily + weekly cycle The Brain 6 operating files The Skills 58 abilities The Hands your tools reads context reaches into logs & updates

The Brain

Your six operating files. Identity. Priorities. Decisions. Clients. Daily anchors. Weekly commitments. What's true right now, not last quarter. The brain needs feeding: three lines every time something real happens.

The Skills

58 abilities. Write the email. Prep the brief. Draft the proposal. Log the decision. Each skill reads the brain before it writes anything. You say what you need. The right skill fires.

The Hands

The tools the skills reach for. Your calendar. Your inbox. Your notes. Your transcriber. The OS doesn't replace them. It connects them. One place to store. One place to talk. One place that holds your time.

The Heartbeat

The rhythm wrapped around all three. A daily anchor in the morning. A weekly retro on Sunday or Monday. Two moments where the system surfaces what's stale, ignored, or parked too long. Without the heartbeat, the brain goes stale and the skills stop firing clean.

Where Founder OS actually specialises.

Most founders don't realise they only need about five percent AI in the right places to operate cleanly. The work moves between humans and skills in a specific order. Humans hold judgment, taste, and the relationships. Skills hold context and produce drafts. Skip the order and the output is generic. Hold the order and one operator does the work of three.

Where to start: tune three skills first

Forty-eight skills is too many to set up cold. Pick three that match what you actually ship in the next seven days. Tune those. Ignore the rest until you need them. This is the difference between people who use the OS daily by week two and people who give up because they tried to perfect everything.

Client emails this week
voice-interviewemail-drafteryour-voice
Reads your voice profile so emails sound like you, not like AI. Runs an anti-AI filter against your captured anti-patterns before returning a draft, so generic phrasing gets caught before send. Leans on your brain snapshot so the draft matches what you're actually working on this week.
Writing an SOP
sop-writervoice-interview
Reads your stack file so 'CRM' becomes the tool you actually use, not a placeholder. Inherits your voice rhythm so steps don't read like a corporate template. Owner and tools fields stay populated so a delegated reader can act without coming back to ask.
A new proposal
brand-interviewproposal-writeryour-deliverable-template
Inherits the colours, fonts, and logo you set during the brand interview. Applies your voice profile so the prose doesn't read like AI. Reads the existing client context file so you don't re-explain who they are or what was last agreed.
LinkedIn content rhythm
voice-interviewlinkedin-postcontent-repurposer
Treats the first two lines as a unit because that's the 'see more' cutoff on mobile. Anti-AI filter applies to every draft. content-repurposer takes one piece and adapts it across LinkedIn, email, and other formats so you don't write five things.
Meeting prep and capture
meeting-prepcapture-meetingknowledge-capture
meeting-prep pulls client history and prior commitments before the call. /capture-meeting routes the post-call dump into log.md, clients.md, and open commitments in one step. knowledge-capture writes takeaways with stable IDs so you can cite them later.
Decision overload
decision-frameworkforcing-questionsbrain-pass
decision-framework adapts to whether you decide by gut, data, or dialogue and runs the right template. forcing-questions runs six fixed questions that test done state, smallest ship, replacement cost, urgency, and proof. brain-pass synthesises across your log, decisions, flags, and patterns so the answer cites your own evidence.
Daily ops triage
todaypriority-triageweekly-review
today gives a 20-line one-screen view: anchor task, open flags, last three log entries, next event. priority-triage cuts to ONE thing, not a list. weekly-review keeps the cadence honest so flags don't pile up.
You'll add more skills as you need them. The OS does not punish you for ignoring 42 of them on day one. Start where the work is.

All 58 skills

Every skill below fires from a plain-language request - no commands to memorize. Click a name to read the exact instructions it follows on GitHub.

approval-gatesCheck whether an action needs approval before doing it.

Why: Prevents irreversible or public actions from happening silently - sends, pushes, deletes, and payments need a human yes first.

Enhance: Fill rules/approval-gates.md with your own thresholds - the default table is a fallback; your custom gates are what actually protect your workflow.

Read full skill →

auditAudit the OS in one composite report.

Why: Gives you one honest number for OS health so you know what to fix instead of guessing which layer is causing bad output.

Enhance: Run after a sprint roll so cadence files are fresh - stale daily-anchors and empty brain log are the two checks most likely to drag your score down.

Read full skill →

blind-spot-reviewPressure-test a draft for blind spots before pre-send review.

Why: Catches the assumptions and missing context that first-draft writers always miss - the categories no one thinks to check until after a deal falls apart.

Enhance: Provide the source material the draft was built from - the review is weakest when run on a deliverable alone and sharpest when it can compare draft claims against meeting notes or brief.

Read full skill →

bottleneck-diagnosticFind the real bottleneck in the business.

Why: Names the one dimension where your presence is most costly so you can stop guessing and start shedding the right role first.

Enhance: Answer all five dimension questions honestly in self-assessment mode - the score is only as accurate as the examples you give for each dimension.

Read full skill →

brain-logCapture a thought, flag, or decision into the brain system.

Why: Keeps thoughts, decisions, and flags from living only in your head - a running log is how the OS learns what is actually happening in your week.

Enhance: Log in Mode B or C when you can - thoughts cross-referenced to a client, decision, or priority are far more useful to skills like weekly-review and brain-pass than standalone entries.

Read full skill →

brain-passSynthesise an answer across the brain layer.

Why: Synthesises an answer from across your brain files with citations rather than asking you to remember what you already logged or decided.

Enhance: Fill brain/log.md consistently with dated entries - the pass is only as good as the history it can reason across; sparse logs return low-confidence answers.

Read full skill →

brain-snapshotRefresh the runtime context payload other skills read at task time.

Why: Gives output-producing skills a cheap read of what is true right now - without it each skill starts cold and ignores open flags or this week's commitments.

Enhance: Re-run after any weekly review, flag update, or major decision so the snapshot reflects current state - stale snapshots silently skew every skill that reads them.

Read full skill →

brand-interviewSet up the visual brand profile.

Why: Creates the visual identity file that every branded document inherits - without it proposals, CVs, and decks default to generic styling that does not look like you.

Enhance: Have your logo file and hex codes ready before starting - the interview can suggest defaults but real brand assets in core/brand-assets/ are what make output look professional.

Read full skill →

brand-voice-interviewCapture a brand's writing voice and positioning.

Why: An operator can run multiple brands. The operator's personal voice and a brand's voice are not the same. Without a brand voice layer, every output sounds like the operator, which is wrong for any branded communication (customer comms, brand social, customer service, ads). Brand voice is the file every brand-coupled writing skill reads.

Enhance: Capture brand voice AFTER you have 3 to 5 real brand samples - existing captions, product copy, customer emails the brand has actually sent. The samples are the ground truth. A brand voice profile without samples is a stereotype.

Read full skill →

business-context-loaderLoad the founder's business context (ICP, offer, pricing) into the session.

Why: Fills in the ICP, offer, and pricing context that proposal-writer and strategic-analysis need - without it those skills produce generic output not tied to your actual business.

Enhance: Walk sections 1-7 in order at least once per business; ICP and offer (sections 4 and 5) have the highest downstream impact on any output skill that reads this file.

Read full skill →

campaign-from-themeTurn one theme into a structured marketing campaign.

Why: Industry-standard campaign generators output a calendar without knowing who the campaign is for, where the audience is in the funnel, or why one piece comes before another. That produces slop that the operator has to throw away. The gate is the value - it forces audience and objective clarity before any draft, which is what makes the output usable on first attempt.

Enhance: Run brand-voice-interview for the relevant brand first - a campaign in operator voice for a brand the operator runs will sound off-brand. The skill works either way but produces sharper output when the brand layer is captured.

Read full skill →

client-updateWrite a client-facing status update or milestone report.

Why: Keeps clients informed without the founder writing from scratch each time - consistent format builds trust and prevents the relationship from going silent.

Enhance: Fill core/voice-profile.yml first so the update sounds like you, and keep context/clients.md current with milestone notes so the skill has real project specifics to draw on.

Read full skill →

content-repurposerReformat one piece of content across many channels.

Why: Multiplies one piece of thinking across multiple channels without re-writing from scratch for each - saves hours on content execution.

Enhance: Fill core/voice-profile.yml before repurposing - without it the output adapts format but loses your voice, which is what makes the same insight work across platforms.

Read full skill →

context-persistencePull what we already know about a topic before re-explaining.

Why: Prevents the OS from asking you to repeat context it already has - skills should read your files before asking you a question whose answer is already written down.

Enhance: Keep context/clients.md and context/decisions.md updated after every interaction - the more accurate these files are, the less the OS needs to ask you.

Read full skill →

data-securityCheck whether data is safe to share with an external tool.

Why: Stops client and sensitive data from moving to tools that were not approved - a single paste to the wrong AI can breach client trust and violate data agreements.

Enhance: Fill rules/data-handling.md with your own per-tool decisions and any client-specific restrictions from context/clients.md - the fallback matrix is generic and may be too permissive for your situation.

Read full skill →

decision-frameworkWalk through a structured decision when the founder is stuck.

Why: Breaks the loop of going around the same decision repeatedly by structuring what you are actually choosing and what each option costs.

Enhance: Fill core/identity.md Decision style field (gut/data/dialogue/mixed) so the skill leads with your preferred mode - a mismatched opener means you disengage before the analysis starts.

Read full skill →

email-drafterDraft an email in the founder's voice.

Why: Writes emails that sound like you sent them instead of like a marketing tool did - especially useful for cold outreach and difficult conversations where tone matters.

Enhance: Fill core/voice-profile.yml and set stack.json email_platform so the skill mirrors your actual send environment and applies your exact voice rules to every draft.

Read full skill →

forcing-questionsPressure-test a new initiative before starting it.

Why: Stops new initiatives from starting before the problem, smallest version, and proof are clear - most wasted build time begins with an idea that cannot answer these six questions.

Enhance: Keep context/priorities.md current so the skill can compare this initiative against what is already in flight and surface real conflicts rather than generic caution.

Read full skill →

founder-coachingCoach a founder through their state and capacity.

Why: Diagnoses where a founder's time and energy is actually going versus where it should go - the session structure and bias toolkit surface what self-reflection alone misses.

Enhance: Run prior session captures in brain/log.md before a coaching session so the skill can reference what the founder committed to last time and surface unresolved patterns.

Read full skill →

founder-os-setupSet up Founder OS from scratch.

Why: Creates the full operating environment in one guided session so skills have the files they need to produce personalized output from day one instead of returning generic defaults.

Enhance: Answer the positioning questions in Phase 0.2.5 as specifically as possible - who you sell to, what you sell, and the buyer pain feed directly into proposal-writer, linkedin-post, and client-update.

Read full skill →

handoff-protocolBuild a handoff doc when work moves to someone else.

Why: Turns a verbal delegation into a structured artifact the receiver can act on without a follow-up call - context dropped in chat alone disappears.

Enhance: Provide all nine required fields explicitly, especially the definition of done and known constraints - the artifact is only as useful as the specificity of what you give it.

Read full skill →

ingestFile a source (URL, file path, or pasted text) into raw/ with provenance preserved.

Why: Preserves a source with provenance intact so you can cite where an idea came from months later - knowledge-capture organises takeaways but ingest keeps the original.

Enhance: Approve 2 to 4 wiki updates per ingest rather than all or none - selectively cross-referencing sources into decisions.md and brain/patterns.md builds a connected knowledge graph over time.

Read full skill →

knowledge-captureCapture takeaways from a book, podcast, article, or conversation.

Why: Turns a book, podcast, or conversation into a retrievable note that skills like proposal-writer and strategic-analysis can cite - learning that stays in your head does not improve your output.

Enhance: Always fill the Apply To sections with real business context - notes with no application mapping are rarely surfaced by brain-pass or strategic-analysis when you need them.

Read full skill →

legal-complianceJurisdiction-aware legal, regulatory, and compliance skill for founders.

Why: Grounds legal guidance in loaded reference documents rather than model memory - a jurisdiction without loaded files gets a refusal rather than a hallucinated answer.

Enhance: Run /founder-os:legal-setup to populate your jurisdiction folder and context/compliance.md with your own deadlines - the skill is only as accurate as the reference docs wired in.

Read full skill →

linkedin-postWrite a LinkedIn post in the founder's voice.

Why: Produces posts that sound like you wrote them rather than like a content tool did - voice profile plus brain context means the post is grounded in your actual situation this week.

Enhance: Fill brain/log.md regularly so brain-pass can suggest angles you have not posted recently and flag theme repetition before you draft - a rich log means fresher, non-repetitive content.

Read full skill →

lintAudit the wiki for integrity issues.

Why: Surfaces broken cross-references and stale content that silently mislead skills - a wiki that points to missing files or a decisions list that contradicts the sprint is worse than no wiki.

Enhance: Run wiki-build first to refresh brain/relations.yaml so lint has an up-to-date graph to check against - lint on a stale graph produces false orphan counts.

Read full skill →

meeting-prepPrep for a meeting or debrief one after.

Why: Builds a specific brief from your own prior context rather than a generic one - knowing what you already committed to and what is unresolved with this person is what makes the meeting productive.

Enhance: Keep context/clients.md updated after each touchpoint so brain-pass can surface real prior commitments - a brief built from stale client data will miss the most important watch-for items.

Read full skill →

menuShow what FounderOS can do right now.

Why: Shows what the OS can do right now based on your actual state - a static skill list tells you what exists; menu tells you what is relevant today.

Enhance: Fill cadence/weekly-commitments.md and keep brain/flags.md current - menu scores suggestions against your must-do list and open flags, so empty cadence files produce a generic starter list.

Read full skill →

observation-rollupRoll up weekly observation logs.

Why: Prevents the brain/observations/ folder from accumulating thousands of raw JSONL lines by compressing old weeks into readable summaries you can actually review.

Enhance: Set FOUNDER_OS_OBSERVATIONS=1 in your shell to generate JSONL files in the first place - the rollup has nothing to process if observation logging is off.

Read full skill →

pre-send-checkRun a pre-send gate before any deliverable goes out.

Why: Forces a structured check before anything goes out so token-replacement misses, broken links, and voice inconsistencies are caught rather than sent to clients.

Enhance: Always provide the source of truth the deliverable was built from - Check 2 (source-truth match) is the most failure-prone check and only works when the skill can compare the draft against real source material.

Read full skill →

priority-triageCut through overwhelm to ONE thing to do next.

Why: Cuts a long open list down to one actionable task by running four filters - the founder who is overwhelmed needs a decision, not another framework to think through.

Enhance: Keep brain/flags.md updated with stall patterns and cadence/weekly-commitments.md current - triage is sharpest when it can see which items have already rolled multiple weeks.

Read full skill →

proposal-writerWrite a consulting or service proposal.

Why: Builds proposals that reflect the specific prospect's situation rather than a generic template - a proposal that mirrors the client's own words wins more than one that describes the service.

Enhance: Fill context/clients.md with scoping notes before running - the more prospect-specific context you provide, the stronger the Situation section, which is where most proposals lose.

Read full skill →

querySearch across the OS with a multi-hop query.

Why: Retrieves OS nodes across files without loading everything into context - start broad with index mode, then drill into the specific entries that matter.

Enhance: Run wiki-build to populate brain/relations.yaml before querying - the graph edges improve candidate ranking and the timeline mode relies on the graph to find connected entries.

Read full skill →

queueManage the work queue.

Why: Enforces a hard cap of three active items so work in progress reflects actual human capacity rather than an aspirational list that never clears.

Enhance: Use the start operation consistently rather than just reading the queue - the 3-item gate only forces the right decisions when you route new work through start instead of adding directly to ACTIVE.

Read full skill →

readiness-checkScore the OS install across five buckets and name the next 3 moves.

Why: Shows exactly which files and buckets are holding your score down so you can fix the right things rather than assuming setup is done when it is not.

Enhance: Run voice-interview and brand-interview first - Voice and Brand carries 25% weight each and is the single fastest way to move the score once Core files exist.

Read full skill →

review-responderDraft replies to incoming customer messages, reviews, DMs, and inquiries.

Why: Incoming customer messages are daily, often time-sensitive, and reveal as much about the brand as outbound marketing does. A reply in the wrong voice (operator personal voice instead of brand voice, generic AI voice instead of either) erodes trust. This skill makes the right-voice reply the default rather than the careful effort.

Enhance: Run brand-voice-interview for the brand whose channel the message came from - replies in the wrong brand voice are more damaging than no reply because they signal the brand is inconsistent.

Read full skill →

session-handoffPackage this session for a fresh chat.

Why: Makes work transferable to a new session or another person without losing the reasoning behind decisions - context in chat alone is gone the moment the session ends.

Enhance: Be explicit about handoff type (A/B/C) at the start - the skill strips internal strategy and inlines skill-dependent knowledge differently for each type, and guessing wrong produces a document that leaks or lacks context.

Read full skill →

ship-deliverableRun the final ship gate before any external deliverable goes out.

Why: Composes four checks that are easy to skip individually under deadline pressure into one pass so nothing ships with an AI phrase, missing asset, or unreviewed blind spot.

Enhance: Run blind-spot-review before calling ship-deliverable - Link 3 (blind-spot evidence check) will fail if no review artifact exists, and you will have to run it separately anyway.

Read full skill →

sop-writerWrite a delegation-ready SOP.

Why: Documents a process so it can run without the founder in the room - the goal is a document a smart person who has never done this task can follow to completion.

Enhance: Fill stack.json with your actual tools before running so TOOLS NEEDED names your real CRM, automation platform, and email client rather than generic labels the delegate cannot look up.

Read full skill →

strategic-analysisRun a strategic analysis: competitive map, market sizing, opportunity, or business model.

Why: Grounds competitive and market analysis in your actual position and prior knowledge rather than producing a generic table that ignores what you already know or have decided.

Enhance: Populate brain/knowledge/ with notes from books and conversations about the market and competitors - the skill reads matching topics by name and uses them as prior evidence in the analysis.

Read full skill →

todayShow today's one-screen brief: anchor task, open flags, last 3 log entries, and next event.

Why: Gives you a one-screen brief of what matters today without loading every context file - anchor, flags, recent log, and next calendar event in under 20 lines.

Enhance: Keep cadence/daily-anchors.md rolled to today and log at least one entry to brain/log.md during the day - a stale anchor and empty log make the brief generic rather than specific.

Read full skill →

unit-economicsRun business math: unit economics, pricing, CAC, LTV, margins, break-even.

Why: Runs business math in a structured format that shows assumptions and scenarios rather than a single number that can be misread without context.

Enhance: Give real numbers rather than ranges when you can - the skill flags assumptions that could change the answer, but it cannot flag the ones you did not provide.

Read full skill →

verifyCheck that Founder OS is healthy.

Why: Checks that the OS substrate is actually wired up correctly rather than just declaring setup done - hooks, scripts, and counts that disagree silently break skills.

Enhance: Run after every setup or update to catch wiring issues early - a FAIL on scripts or hooks means several skills will behave incorrectly on every subsequent run.

Read full skill →

voice-interviewSet up the writing-voice profile.

Why: Builds the voice profile that gates every writing skill - without it linkedin-post, email-drafter, proposal-writer, and client-update all produce generic output that does not sound like you.

Enhance: Paste at least three real samples you have already written before answering shaping questions - samples are the ground truth and override the rules when they conflict, so more specific samples mean a sharper profile.

Read full skill →

weekly-reviewRun the weekly review and roll the sprint.

Why: Keeps the sprint honest by forcing a decision on stalled items instead of letting them roll week to week - the pattern block names avoidance directly.

Enhance: Fill brain/log.md throughout the week with real events - the retro reads from this directly, so more entries means a sharper pattern read and a more honest retro.

Read full skill →

wiki-buildRebuild the wiki graph from extracted [[wikilinks]].

Why: Keeps the wiki graph current after a session that added cross-references so lint and query have accurate edges to work with - without it the graph drifts from the actual files.

Enhance: Run after any session that adds or changes [[wikilinks]] in operating files - the script is idempotent so running it twice does no harm; the cost of a stale graph is broken lint results.

Read full skill →

your-deliverable-templateBuild a branded document inheriting the founder's colors, fonts, and logo.

Why: Ensures every document that carries your name inherits your visual identity rather than defaulting to generic black-and-white formatting that does not represent you.

Enhance: Place logo files at the paths recorded in core/brand-profile.yml before generating any HTML or PDF - the skill checks for real files and stops rather than silently substituting a placeholder.

Read full skill →

your-voiceApply the founder's writing voice to any text.

Why: Applies the captured voice profile to any text so every output sounds like you rather than like a language model - it is the layer that makes personalization real rather than cosmetic.

Enhance: Add anti-example pairs to core/voice-profile.yml over time by flagging outputs that do not sound like you - each pair tightens the profile and reduces future corrections.

Read full skill →

The Hands - one place for each

If skills are the abilities, your tools are the hands. The hands hold things, send things, store things, listen. The OS doesn't replace your inbox or your calendar - it connects what skills need to do with the tools that already do it. One place to store. One place to talk. One place that holds your time. You choose the tools; the OS connects.

Two automations to ship first

Calendar booking. People pick a time. Your calendar knows. The OS knows. Transcription. Every call captured, no notes during the conversation. Together: who you spoke to + what you spoke about, in the brain, without typing a word.

Storage

One place for what you've written, decided, agreed on. The knowledge base skills read from when context matters.

Notion - or Coda, Obsidian, Google Docs
Calendar

One place that knows when you're booked. Meeting-prep, weekly-review, and today read it to surface what's actually coming.

Google Calendar - or Outlook, Apple Calendar
Email

Where async with the outside world lives. Skills draft from thread context; you approve before send.

Gmail - or Outlook, Fastmail, Proton
Communication

One channel for your team and clients. Skills read context, draft updates; sending stays manual unless you explicitly authorize it.

Slack - or Telegram (easy API). WhatsApp works too, but only via the Business API.
Booking

External people pick a time without back-and-forth. The booking writes to your calendar; the OS sees it.

Calendly - or Cal.com, SavvyCal
Transcription

Capture conversations without note-taking. Capture-meeting parses the transcript into log, decisions, and follow-ups.

Granola - or Otter, Fireflies, Read.ai
We don't pick favorites. Notion or Coda. Slack or Telegram. Google or Microsoft. Connect whichever you already use. The skill connects, not the other way around. The OS adapts to your stack, not the reverse.

Each tool connects once - a short setup, no code. Behind the scenes it's an MCP, the standard Claude uses to talk to outside tools. Nothing routes through ARCAS. You own the connection. You own the data.

Seven days, not seven hours

Don't try to do this in one afternoon. The pattern that kills every system: you try to install it all at once. Your brain glazes over. By Tuesday you've stopped.

And don't try to map where the OS will work and where it won't before you've used it. You'll know more after a week of living in the system than after a month of planning it. Set it up. Use it on real work. The right places to lean on it become obvious from the inside.

Here's what the first seven days look like instead. Click each day to expand.

1
Install and your first six files
30 minutes
Run the setup wizard. Answer the prompts. The system generates your identity file, priorities, cadence, brain layer. Don't add anything extra. This step alone gets daily ops, meeting prep, decision logging, and knowledge capture working.
2
Your voice and your brand
20 minutes
Run the voice interview. Run the brand interview. From this point, every writing skill produces output that sounds like you - not like AI wrote it. This is the step most people skip and then wonder why their emails feel off.
3
Log one real thing
5 minutes
Something actually happened today. A decision. A conversation that changed something. A deal that moved. Log it. Three lines in your daily log. That's the muscle you're building. The first log entry is the hardest. After that it's a habit.
4
Run your first rant
10 minutes
Day 4 is when you first dump an unstructured thought into the system. No editing. No structure. Use /rant. Just talk. This is how you start capturing what's actually in your head, not just the cleaned-up version.
5
Use one skill on real work
15 minutes
Not a test. Something you actually need. A meeting prep. An email draft. A LinkedIn post. Let the skill read your brain and produce output. See what comes back. Adjust one thing. This is how the system starts feeling useful, not just set up.
6
Process the rant
10 minutes
Run /dream on what you captured on Day 4. The skill turns your raw thinking into flags, patterns, and parked decisions. This is how the brain learns what you care about. The rant was raw material. Dream gives it shape.
7
Run the weekly retro
20 minutes
Look at what you said the week would be. What actually happened. What moves forward. What gets killed. What decision you've been avoiding. This is the heartbeat. Once it's running, it runs. Put it in your calendar. Sunday at 9pm or Monday at 8am.
Perfect doesn't ship. Working compounds. By the end of seven days, you have a working system. That's the goal - not a complete one.

What day 30 looks like

This is a real session. Not a demo. This is what the system looks like when it's working.

8:52 am

You open Claude Code. The session brief appears before you've finished your coffee. Two flags open from last week. One decision overdue since yesterday. The Ahmed proposal has been sitting for three days.

You didn't read any file to find this. The system read them.

9:06 am - prep for the 10am call

You type: "prep me for my 10am with Ahmed."

Four minutes later you have a brief. It knows Ahmed runs a 22-person firm in Abu Dhabi. It knows what you discussed on the last call. The staffing model question he was wrestling with. What you said you'd follow up on. The two things he flagged as blockers.

You wrote none of this today. You logged it three weeks ago. Three lines. After a different call.

11:02 am - after the call

You type three sentences.

Ahmed confirmed budget. Moving to proposal stage.

Wants phased rollout, not all at once.

Proposal due Thursday.

That's it. Next time you open Claude Code, those three sentences shape every skill that touches Ahmed. The proposal skill knows about the phased rollout. The client update skill knows Thursday. You don't explain again.

End of day

Eleven minutes of admin across the whole day. Two at the start. Four on the prep. Five on the capture. The system didn't replace the call or make the decisions. It just meant you showed up knowing what mattered, and left without anything slipping.

Day 90 looks the same. The output is sharper because the context is richer.

The language of AI

AI reads structured text faster and more accurately than prose. Headers, bullet points, bold text, numbered lists give the model clear signals about what matters. Write a wall of unstructured text, the model reads a wall. Write with structure, it works with structure.

That's why your pages use markdown - the format AI was trained on most. You never have to learn the syntax. The setup wizard writes it for you. You just type sentences.

# priorities.md ## This week - Close Maya deal AI model Output in your voice Ready to send to Maya by Thursday

This is why every page in your OS is plain text. Not because it looks nice in a text editor. Because it gives the model the clearest signal about what matters - and gives you a file you can open in any tool, on any machine, and read in twenty seconds.

The model is trained on the world. It doesn't know your world until you build the files. Every piece of context you add - a new client, a decision, a standard you hold - the model gets more specific. The output gets more useful. As the models improve, that context compounds. Same honest files, better model, better output. You don't start over when there's a new model. Your files stay.

identity .md priorities .md decisions .md clients .md weekly commit .md daily anchors .md

Your brain files as a graph view - how Obsidian renders the connections between your markdown files. Every line is a [[wikilink]].

Obsidian - obsidian.md
Markdown editor with graph view. Mac, Windows, Linux, iPhone, Android. Free for personal use.
The jobRead your OS files as a connected graph, not a folder of documents.
Why this matters even without AIMarkdown is easier to navigate when you can see how files link to each other. You spot orphans, dense clusters, and missing connections at a glance.
Why it compounds with the OSThe OS uses [[wikilinks]] between brain files. Obsidian renders them as a graph. Adding one well-placed file improves the output of six different skills at once - the graph view is how you see that.
Why we picked this oneObsidian opens your repo folder directly. No import. No conversion. Your files stay where they are.
You can pick anotherLogseq, VS Code with a markdown extension, or any text editor reads the same files. The graph view is the differentiator.

Your files are the memory, not the session

Every AI session starts fresh. The model doesn't remember what you said last Tuesday. When you close the window, that conversation is gone.

This is why people get frustrated with AI assistants. They re-explain everything every time. They spend five minutes catching the AI up on what it forgot.

Session memory You talk to Claude → context builds → window closes → GONE File memory (this OS) Files live on your machine → model reads them every session → window closes → FILES STILL THERE ✓
If you've tried AI assistants before and they didn't stick, here's exactly why. Every session started from scratch. The AI had no memory of your business, your standards, or your relationships. It wasn't the AI that failed. It was the missing context layer. That's what this OS provides.

Founder OS solves this by making context live in files, not in the session. The model reads the files every session. Fresh read, same truth. You don't re-explain. The files explain.

And because the context is in files, you're not locked to any one tool. Claude Code today, Codex tomorrow - they both read the same six files. The OS ships with an AGENTS.md file for exactly this reason. Non-Claude agents pick up the same operating rules.

Data is the real leverage.

In Founder OS, you are the data. Your decisions, your rants, your client notes, your stalled flags. The model is the same model everyone else uses. The difference is what you've fed it. The longer you run, the more your file layer compounds, and the harder it gets for a generic chat to match what you get out of yours.

Local-first means: the data lives on your machine. Not on a company's server. Not in someone's cloud. Plain markdown. Any text editor reads it. Any AI can load it. You own it.

You could use any white-label tool out there. The trade is real: you're locked to that vendor for as long as the business runs. Your team was your dependency before. Another tool would just shift it. Founder OS is your spreadsheet. Save it, import it into Google or Excel, drop it into a different AI - doesn't matter. It belongs to you before it belongs to any tool.

Capture meetings before you forget them

Meetings are where decisions are made. Where commitments are given. Where relationships shift. But if you don't capture them, they might as well not have happened. You walk out with three important things in your head. By the next meeting you remember one.

Granola - granola.ai
AI notepad for back-to-back meetings. Mac, Windows, and iPhone. Freemium.
The jobCapture what was said in a meeting without having to retype it after.
Why this matters even without AIYou stop losing 20 minutes after every call writing up notes. The transcript is on disk. You can find it later. That alone is the value.
Why it compounds with the OSPaste the summary into /capture-meeting. The OS routes decisions into clients.md, commitments into the log, and follow-ups into open work. You walk into the next call already prepped.
Why we picked this oneGranola transcribes computer audio directly. No bot joining the call. No third party in the meeting. Works for back-to-back calls without setup.
You can pick anotherOtter, Fireflies, Tactiq, and Fathom all do this. We are not selling Granola. The point is that a transcriber exists at all.

Set the transcriber running before you start the call. Let it work in the background. Review the summary before you close the laptop. Then run /capture-meeting. That's the whole loop.

Rants

Humans don't think in bullet points. We think in rants.

Most of your best thinking happens while you're walking, driving, or half-asleep. Not when you're sitting in front of a keyboard. A rant is how you capture the real version of that thinking before it's gone.

A rant is an unstructured voice dump of what's in your head. No format. No cleanup. You don't need to know what you're trying to say before you say it. You just start talking.

/rant Raw thinking brain/rants/ Stored raw /dream Distilled Flags Patterns Decisions

The /rant command opens a capture. You dump. The system takes it raw and stores it in brain/rants/. Later, you run /dream. The dream skill reads the rant and distills it: patterns noticed, flags worth tracking, decisions you were circling, open questions.

Rant captures the mess. Dream gives it shape.

Dream's output is also a prompt you can paste into another chat. If you're going to do the heavy work in a different tool - Cloud Claude on your phone, ChatGPT, Gemini, anything - the rant-to-dream loop is how you arrive at a prompt that already knows your context.

Rants are where you think. Prompts are where you ask. By the time you type a prompt, you should know what you're asking. The model is not there to do your thinking. The best prompts are the honest ones.

Talk to your system

If you're going to rant, use your voice. The friction of typing is exactly what kills the habit. When the gap between thought and capture is too big, you let thoughts go.

Wispr Flow - wisprflow.ai
Voice dictation that types anywhere on screen. Mac, Windows, iPhone, and Android. Free trial, then paid.
The jobSpeak instead of typing. Anywhere on screen.
Why this matters even without AIYou speak more of what you actually think than you type. Typing edits as you go. Voice does not. Your unfiltered thinking gets captured before you tidy it.
Why it compounds with the OSRants are how the OS captures the messy version of your thinking. Voice is what makes rants real. A typed rant is a draft. A spoken one is the truth.
Why we picked this oneWispr Flow types straight into any field on any app. No copy-paste. The friction between thinking and capture drops to zero.
You can pick anothermacOS Dictation and Windows Speech Recognition are free and built in. Phone voice memos plus paste also work. The point is voice, not the tool.

Talk to the system like you're thinking out loud. Look at what came out. Edit it if you want. Paste it in. Speak, look back, refine. That process is how you get better prompts without spending 20 minutes composing a perfect question.

The best prompts are the honest ones.

Skills work in Claude too

Skills are markdown files. They are instructions Claude follows. That means they are not Claude Code-specific.

Claude on web, desktop, and mobile can read skills inside Projects. Tune your voice-interview, email-drafter, or proposal-writer skill in your local repo, then use that same skill file in a Claude Project. Bring only the relevant context with it. The source of truth stays on your machine.

PLAIN CHAT
Chat
Cowork
Code
+ New chat
Projects
History
NO PROJECT CONTEXT
A clean chat only knows what you paste.
Can you help me write this?
No Founder OS folder attached
Plain Claude chat. Useful, but generic.
WITH YOUR FOLDER
Chat
Cowork
Code
+ New task
Projects
Files
FOUNDEROS FOLDER OPEN
Prepare your second brain.
Can you help me set up my second brain?
Downloads\FounderOS-main
Claude Code: best start Cowork: optional
Open the folder, then say: Start setting up.

Cloud chat alone is the receptionist - it answers, logs somewhere, polite, doesn't know your context. Drop your skills folder in and type start - it becomes a personal executive assistant. Tune the voice and brand files - it becomes a specialist that can actually do the task. Same chat. Different context.

Three concrete moves you'll use this for:

Skills are portable. Your repo is the source of truth, but the skills travel with you.

What stays local: hooks, scheduled tasks, file writes, and the session-start brief are Claude Code-only. Skills work everywhere. The rest of the OS runs on your machine.

One thing to know about Cloud Claude: reading a whole folder needs GitHub. Connect your repo to a Claude Project and the skills load automatically. No GitHub? Two clean ways through: (1) upload the specific skill files you need into the Project, or (2) install the Claude desktop app and use that with your local folder. The OS files travel either way - the connection method is the only thing that changes.

Working with a team

If you have people around you, the OS adapts. But the setup changes.

The OS as described here is personal. The files hold your thinking, your relationships, your unfiltered decisions. That's not something you share wholesale.

What you share with a team is outputs, not brains.

When a skill produces a proposal, a meeting summary, an SOP, or a client update - that's a clean deliverable. It's ready to share. The OS has a pre-send check (/ship-deliverable) that runs before anything goes to a client. It verifies the output matches your voice, your standards, and the facts in your files.

For team members who want their own OS, they install separately. Same structure, different truth. Two operating layers that communicate through deliverables, not through shared files. Each person holds their own context. The outputs connect the work.

Can my team use Founder OS?

Yes. They install separately. They answer their own setup questions. They tailor the skills for how they actually work. Two operating layers that share your standards and your brand, not your inbox. You keep your context. They keep theirs. The deliverables connect the work.

What not to put in

The OS is local. The files live on your machine. But they're also in git, and they sync to GitHub if you push. Know what that means.

Goes in

Priorities, clients, decisions, notes, flags, contact names, general business context. The brain should hold enough for the AI to be genuinely useful.

Stays out

Passwords. API keys. Bank details. Full government ID numbers. Anything a third party sent you in confidence. Legal advice you haven't reviewed. Anything you wouldn't want in a public document if the repo leaked.

The system has a privacy pre-commit hook. Anything you wrap in <private> tags gets stripped before it pushes to GitHub. Use it for anything borderline. But the cleaner practice is to not put sensitive data in the files at all.
On data: your files live on your machine. When you run a session, the OS includes them in your conversation context. That context is processed by your AI provider during the session. Check their privacy policy before adding sensitive business information to prompts.

The OS is a thinking layer, not a data warehouse. It should hold enough context to be useful and nothing that creates risk if it ends up somewhere unexpected.

When AI gets it wrong

It will. Plan for it.

The model is trained on the world. When it doesn't have enough context about your specific situation, it fills the gap with educated guesses. Confident ones. Sometimes wrong ones.

When output feels generic: the input was generic. Feed the brain more.

When output feels confidently wrong: push back. Ask the model to show its reasoning. A good model will revise with reasons. A bad one will just agree.

This is the trap. The model will agree with you if you push. It adapts to your framing. That's useful when refining output. It's dangerous when you want honest analysis. If you go in wanting validation, the system will validate you.

The forcing-questions skill exists for this reason. It asks six questions that have a real chance of killing the idea. That's the point. Better to kill a bad idea in six questions than spend three weeks building it with AI helping you go faster.

Don't ship AI output you haven't read. Every deliverable that leaves your machine has your name on it. The pre-send check (/ship-deliverable) is the built-in gate before anything goes to a client.

Some weeks, the team member who took a sick leave was probably better than the AI.

We don't ask you to believe that until you've felt it. Use the OS where it lifts you. Notice where it doesn't. Both signals matter. The goal is to know the difference, not to use AI everywhere.

What can go wrong

Honest list. These are the failure modes.

You stop logging

The system goes stale. Stale context produces generic output. Generic output feels like AI, not like an assistant that knows you. You stop trusting it.

Fix   Three lines per day. Doesn't have to be more.

You fill everything in Week 1

You read the docs. You see 58 skills and 33 commands and a whole brain layer to build. You try to set it all up before you use any of it.

Fix   Use before you optimise. Day 1 through 7. In order.

You forget the cadence

You skip the retro. Then again. Flags pile up. The system starts feeling like something you're behind on. You avoid opening it.

Fix   Put the weekly retro in your calendar. Sunday at 9pm or Monday at 8am. 20 minutes.

You treat it like a chatbot

You ask it questions without feeding it context. You expect generic AI output. You get generic AI output.

Fix   Quality in, quality out. Feed the brain. Use the skills.

You wait for perfect

You want every file filled perfectly before you run a real skill on a real piece of work. The bar keeps moving.

Fix   It will never be perfect. It will be yours. Run the skill. Use the output. Adjust.

The context window drifts

Long sessions on complex topics produce worse output at hour two than minute ten. You don't notice because you're in the flow.

Fix   Start a fresh session after two hours. The files hold context. The session doesn't need to.

The one thing

If you take nothing else from this playbook:

Log what happens.

Not in a diary. Not in paragraphs. Three lines in your daily log every time something real happens. A decision made. A meeting that shifted things. A deal that moved.

The system compounds from that. Skills get sharper. Stall detection gets accurate. Output starts sounding like something an assistant who's been with you for a year would say.

That's the whole game. Not the 58 skills. Not the 33 commands. The log.

When you're ready for more

Founder OS works as a folder on your laptop with no extras. That's the version most founders never need to outgrow. When the friction shows up - and only when it shows up - the same files extend further.

Add these one at a time, only when the friction shows up. Most founders never need more than the laptop folder. The OS doesn't get better because you added a server. It gets better because you fed the brain.

What it costs

One AI subscription. That's it. Everything else is free.

  • Founder OS (this repo) Free
  • Claude Code (the interface) Free
  • AI subscription - Claude Pro or Max

Any plan with 100K+ context tokens works. Free tiers usually don't have enough context. You'll get worse output and sessions that cut off early. A paid plan is the only real requirement.

The OS is built to run inside Claude Code, so a Claude plan (Pro or Max) is the one that fits it end to end. Copilot does not work for this. The files themselves are plain markdown, so you can open them in ChatGPT or Gemini and still get value from the content, but the guided setup and the agent workflows in this playbook assume Claude Code. The files travel with you. You're not locked in.

The subscription is replaceable. The files are yours.

People don't fail processes.

Processes fail people.

Download