🛠️ Operating in Practice

A skill that reads/writes aios-intake.md, conducts the 7-question interview and scaffolds the Day 1 file set at the end. Idempotent: running it again only updates what changed.

Without /onboard, the system stays generic. With it, Claude Code already knows your business, your voice, and your priorities from the first conversation—installing the Default Shift immediately.

Idempotency · combined wizard · aios-intake.md · Default Shift · Day 1 scaffold.

Seven questions with a hard cap: Q1 who/what you sell/to whom · Q2 voice samples · Q3 90-day priorities · Q4 where revenue drops off · Q5 where you talk to customers/the team · Q6 where meetings/docs live · Q7 the task that eats up your week. Each under 60s.

The cap of 7 is intentional: more questions = fewer answers. Seven is the minimum needed to generate enough context and the maximum that won’t wear you out on Day 1.

Hard cap · 7 context domains · revenue · voice · 90-day priorities.

The one unbreakable rule of /onboard: voice samples must be PASTED verbatim, never typed during the conversation. "If you type here while we’re talking, the sample has already been shaped by the conversation."

Voice is what sets generated content apart. A sample contaminated by the session produces text that sounds like "generic AI" instead of sounding like you. /onboard actively refuses if it detects live typing.

Verbatim · voice contamination · references/voice.md · sample vs. voice description.

At the end of the 7 questions, /onboard generates: context/about-me.md, context/about-business.md, context/priorities.md, references/voice.md, connections.md (7 lines) and fill in the CLAUDE.md. Backup at archives/intake-{ts}/.

These files are the system’s permanent memory. Without them, every conversation starts from scratch. With them, Claude knows who you are, what you sell, and where to focus — without you having to explain it all again.

File set · context/ · references/ · CLAUDE.md · backup in archives/ · idempotency.

Automatic prompt at the end of /onboard: "what should I focus on this week?" The answer uses ONLY the generated context files, in 3 bullets in the user's voice, each tied to a 90-day priority.

This is the moment that plants the Default Shift: the user sees the system respond for the first time like a strategic partner, not a typist. The first experience determines whether the ritual will stick.

Default Shift · wow moment · 3 bullets · user’s voice · 90-day priorities.

Edit aios-intake.md and run /onboard again. It refreshes only what changed, skips questions already answered, and automatically backs up before overwriting.

Your business changes. When it does, just edit the intake and rerun it — no need to reconfigure from scratch. Idempotency turns onboarding into a regular update routine.

Rerun · incremental diff · automatic backup · aios-intake.md as the source of truth.

No more than 7 questions · doesn’t write .env (no keys on Day 1 — connections come on Day 2) · does not generate extra skills · read-only on the 3ms-framework.md.

Explicit limits prevent scope creep. /onboard is narrow by design: it does one thing well. Connections and automations come in the following tracks — don't mix them in.

Narrow by design · scope creep · connections on Day 2 · read-only · separation of responsibilities.

/audit scans the system across the 4 Cs (Context, Connections, Capabilities, Cadence), assigns a score to each dimension, identifies gaps, and produces a prioritized list of next actions based on leverage.

Without a diagnosis, improvements are random. /audit makes progress visible—you see the score rise week after week, which reinforces the behavior and justifies the time investment.

4 Cs · health score · gaps · next actions · leverage.

Checks whether context/, references/voice.md, priorities.md, and CLAUDE.md are present, complete, and up to date. The score drops if files are vague, outdated, or missing.

Context is the foundation of everything. A system with excellent Capabilities and Connections but poor Context will produce generic answers. The audit requires reviewing context before scaling.

Context score · completeness · freshness · CLAUDE.md as radar.

Scans connections.md and configured MCPs. The score reflects how many of your business's real tools the system can access versus how many still require manual copy-paste.

Connections are leverage multipliers. Each connected MCP eliminates a manual bridge. The audit shows exactly where the remaining bridges are and prioritizes which ones to break first.

connections.md · MCPs · manual bridges · integration leverage.

Counts and evaluates the skills in .claude/skills/ and artifacts in decisions/log.md. Score measures depth (specialized skills) vs. breadth (areas covered).

Capabilities grow with every /level-up. The audit shows whether you’re accumulating skills strategically or randomly—and identifies areas of the business that still lack coverage.

Skills · artifacts · decisions/log.md · depth vs. breadth · strategic coverage.

Checks whether cadences are defined, which are running without an explicit request, and which still depend on a manual trigger. The score penalizes tasks that "should run on their own" but don't.

Cadence is the hardest C to install. The audit makes the gap between “I automated it” and “it’s actually running” obvious—a difference that would normally take weeks to discover without a diagnosis.

Cadences · manual vs. automatic trigger · execution gap · Cadence score.

At the end, /audit delivers: a score from 0–100 for each C + total · prioritized action list (highest impact first) · suggested next /level-up target.

The scorecard creates accountability. Seeing the number rise from 42 to 67 in two weeks is what turns the operator from “I’m trying” to “I’m making progress.” The list removes choice paralysis.

Score 0-100 · accountability · prioritized list · next target · visible progress.

First run: Day 7 after /onboard. From then on: weekly (you can pair it with /level-up every Friday) or biweekly for anyone in a phase of rapid expansion.

Frequency determines the speed of improvement. Monthly audits = very slow feedback. Daily audits = overhead without a clear signal. Weekly is the proven sweet spot in the AIS-OS kit.

Day 7 · weekly ritual · feedback frequency · overhead vs. signal.

A skill that walks through the 3 Ms (Mindset → Method → Machine) every week to find and ship ONE automation. In addition to delivering the artifact, it instills the 3 Ms in the operator’s mindset.

Most people try to automate everything at once and deliver nothing. /level-up forces you to focus on ONE thing per week — the pace that produces 52 automations/year without burnout.

3 Ms · one automation/week · brain-rewire · structured interview · scaffolded artifact.

5 questions to find candidates: what have you done 3+ times? Anything manual/copy-paste? "Could a smart intern handle it"? What would break if 500 customers showed up? What could serve 500 customers? Result: 1-3 candidates ranked by leverage.

Most people want to automate what’s exciting, not what’s blocking progress. The 5 questions force you to identify the real bottleneck before building anything.

Candidate · leverage ranking · real bottleneck · mental copy-paste · scale bottleneck.

5 steps: Find constraint → EAD (Eliminate first; if "nothing breaks," that's a win—log it) → Map (5 elements) → Autonomy (default to the lowest level) → KPI (bucket + metric, required; otherwise stop).

EAD is the most counterintuitive and most valuable step: sometimes the best "automation" is eliminating the entire task. The required KPI prevents you from building things that can’t be measured—and therefore can’t be improved.

EAD · constraint · Map 5 elements · Autonomy ladder · KPI as an exit gate.

4 options in Boring-is-Beautiful order: (1) saved prompt → (2) deterministic skill → (3) AI-assisted skill → (4) sub-agent (last resort). Default = the biggest option WITHOUT AI that solves it.

The natural tendency is to go straight to the complex sub-agent. Boring-is-Beautiful reverses this: simple solutions are more reliable, easier to debug, and deliver faster. AI is the last resort, not the default.

Boring-is-Beautiful · complexity ladder · saved prompt → skill → sub-agent · simple by default.

Every artifact starts with bike-method-phase: 1 in the frontmatter. This locks the user into Phase 1 (manual validation). Advancing to Phase 2 requires explicitly editing the file — no shortcuts.

Most people skip manual validation and only discover weeks later that the automation doesn’t work. The lock enforces a human feedback cycle before scaling — saving time and preventing production bugs.

bike-method-phase · frontmatter · manual validation · explicit Phase 1 → 2 → 3 · feedback loop.

Each /level-up run produces exactly: 1 entry in the decisions/log.md (Method spec) + 1 scaffolded artifact + 1 closing screen with a Phase 1 reminder.

The output contract ensures traceability. Without a decisions log, you don’t know what was decided or why. With one, in 6 months you’ll have a history of 24+ automations with rationale.

decisions/log.md · spec · scaffolded artifact · traceability · rationale.

After running /level-up every Friday for 4–6 weeks, the 3 Ms become second nature. You start spotting automation opportunities during meetings, emails, and conversations—without needing the skill to see them.

/level-up isn’t just a tool: it’s a cognitive training program. The ultimate goal isn’t the skill itself, but permanently installing the AIOS operator mindset.

Brain-rewire · 3 Ms internalized · operator mindset · 4-6 runs · rituals every Friday.

Immutable base structure: context/ · references/ · decisions/log.md · archives/ · connections.md · .claude/skills/ · aios-intake.md · CLAUDE.md. Removing any one of them breaks the system.

Knowing what not to remove is just as important as knowing what to add. Each file in the foundation has a critical role—without it, an entire skill stops working.

Base structure · core immutability · implicit dependencies · integrity kit.

When the operation justifies it: projects/ (2+ workstreams) · templates/ (stopped copying prompts) · brand-assets/ · references/sops/ · references/{tool}-api.md · scripts/ · .claude/agents/ · sub-OS folders.

Each expansion has a clear trigger (“stopped copying prompts” = templates/). Without this criterion, you create folders in advance—and they become a basement in 2 weeks.

Expansion trigger · templates/ · projects/ · sub-OS · on-demand expansion.

Five fatal anti-patterns: email/Slack dumps in references/ · folder-of-folders (a flat structure with good names wins) · notes/misc/tmp/inbox (graveyards) · pre-creating folders · duplicating decisions · more than 1 CLAUDE.md at the root.

Each anti-pattern turns AIOS into a basement. The difference between a “well-run business” and “digital chaos” is consistently following these 5 hygiene rules.

Anti-patterns · flat structure · digital graveyard · pre-creating · duplication · single CLAUDE.md.

Cadences per file: decisions/log.md (every decision made) · archives/ (quarterly cleanup) · sops/ (when someone new reruns a process) · connections.md (each new tool) · CLAUDE.md (quarterly review).

Without cadence, files become outdated and lose value. With cadence, the system grows organically—each file has a clear owner and schedule.

Cadence per file · immediate decision · quarterly cleanup · periodic review · living system.

Before creating a folder: (1) is it conceptually new? (2) will I use it 3+ times next month? (3) would /level-up route a future skill here? 2 yeses = add it. 1 yes = wait.

It's the gateway to any expansion. Most people create a folder on first use. The rule requires evidence of a pattern before structuring it — preventing 80% of the junk drawers.

3 yeses · expansion gate · pattern before structure · 2/3 yeses = create · 1/3 = wait.

Advanced frameworks for when the system grows: Data Retrieval Hierarchy · Integration Ladder · Error Handling Playbook · Model Selection · Context Engineering · Discovery · Security & Permissions. Core philosophy: "Your AIOS should look like a small, well-run business — not a hoarder's basement."

As the system scales, these frameworks provide patterns for decisions that don’t have an obvious answer. Knowing them before you need them means you won’t have to improvise under pressure.

Branch frameworks · scaling · basement vs. well-run business · decisions under pressure · operating philosophy.