How a 72-line markdown file saved 0.4 months of runway in 15 minutes — and how to install yours today.

I built an AI CFO. He refused me yesterday. He was right.

30 seconds from approving an annual SaaS renewal — vendor portal open, "Approve" button highlighted, brain already on the next task. Then the file said no.

Not "are you sure?" A specific, structured refusal: runway 8.2 months, my policy threshold 12 months, this renewal drops it 0.4 months deeper below threshold. Drafted 3 alternatives. Named the override path. Held the line.

I closed the portal. Called the vendor. Moved annual to quarterly in 2 emails. Saved 0.4 months of runway in 15 minutes of work I almost skipped.

The file is 72 lines of markdown. It lives at c-suite/cfo/SKILL.md in a free repo I open-sourced this week. MIT licensed. No SaaS to subscribe to.

In the next 10 minutes I'm going to show you exactly what's in it, how to install yours, and what to expect in your first week of running it.

What This File Actually Does

The CFO file is one of 11 advisors in myinc-os — an open-source pack of solo founder AI co-founders. Each advisor is a markdown file with persona, operating rules, a refusal list, and 10 numbered skills they own.

The CFO's 10 skills:

1. /runway-snapshot — 13-week cash flow forecast + months of runway + burn trajectory + 3 scenarios
2. /pricing-model — Current state + proposed change → 3 scenarios + breakeven + risk threshold + recommendation
3. /investor-update-draft — Monthly email: metrics + progress + asks + risks + cash position
4. /unit-economics-review — MRR, ARR, ARPU, CAC, LTV, LTV:CAC, payback, gross margin, NRR, Rule of 40
5. /vendor-contract-review — Cost lens: total commitment + alternatives + negotiation levers + cancellation risk
6. /budget-vs-actual — Monthly variance + revised forecast
7. /cash-flow-forecast — 13-week rolling + scenario stress test
8. /board-financial-pack — Quarterly P&L + cash flow + balance sheet + key metrics
9. /cohort-retention-analysis — Customer cohort revenue retention + churn pattern + intervention
10. /pre-spend-check — Any spend over a threshold: gated by ROI window, runway impact, alternatives — refuses to greenlight below 12mo runway

That last one — /pre-spend-check — is the one that saved me 0.4 months of runway on Wednesday. It's the most actionable skill in the entire pack, because it fires in the moment you're about to make a decision you'd otherwise regret.

Here's exactly how it works.

/pre-spend-check — The Refusal That Catches You In The Moment

You invoke it before any spend over a threshold you set. For me, that threshold is anything above a meaningful monthly commitment or any renewal of an existing subscription.

The flow:

/cfo pre-spend-check

CFO INPUT:

• Vendor: \[name\]
• Type: \[new / renewal / expansion\]
• Commitment: \[amount + duration\]
• Trigger: \[founder initiated / scheduled / external\]

CFO ANALYSIS:

• Current runway: \[calculated from latest financial state\]
• Threshold (policy): 12 months
• Runway impact of approval: \[delta\]
• Post-approval runway: \[new state\]

\[REFUSAL ACTIVATED if post-approval runway < policy threshold\]

Reason: Runway is below policy threshold and approval reduces it further.

Override path:

• Founder writes 1-line justification ("strategic, can't be paused")
• Logged to .myinc/overrides.jsonl
• Logged to next /retro

Alternative options:

• Negotiate 30-day pause (vendor often accepts)
• Negotiate quarterly billing (changes effective runway impact)
• Cancel; evaluate replacement at lower tier
• Defer to next quarter

Recommendation:
\[specific to current state\]

The file does three things in sequence:

1. It calculates. Reads your current financial state, projects the impact of the spend, and computes the new runway.
2. It refuses. If the post-approval runway falls below your threshold, it does not greenlight. It does not say "are you sure?" — it says no, with reasons.
3. It proposes. It drafts the alternatives you would have thought of an hour later, while standing in the shower. Quarterly billing. Pause for 30 days. Lower tier. Defer to next quarter.

The override path exists. You can still approve the spend. But you have to write a one-line justification, and that justification gets logged. Two things happen as a result:

• You usually don't bother with the override (the friction kills 80% of bad spends).
• When you do override, the log gives you a record in your next retro of what you decided was strategic enough to break policy for.

This single behavior — refusal + alternatives + logged override — has replaced an entire category of "wait, did I really need to spend that?" remorse from my week.

What's In The File (72 Lines, Walked Through)

Here's what each section does.

Frontmatter (5 lines)

---
name: cfo
description: Cash defender. Pricing modeler. Investor liaison. Refuses spend that drops runway below 12 months without explicit founder override.
layer: 1
paired\_specialists: \[metrics-bot\]
---

The description tells the agent runtime when to load this file. The layer: 1 marks it as a decision-layer advisor (not an executor). The paired\_specialists field links it to Layer 2 specialists that will eventually execute its directives.

Persona (8 lines)

The persona paragraph defines how the CFO thinks. Mine reads (excerpt):

You are the founder's CFO. You think in runway, unit economics, and cash flow. You're skeptical of any spend without a clear ROI window. You speak in numbers and timeframes, not adjectives. Your bias is preservation: when in doubt, conserve cash.

This is the most important section to customize for your business. If you're a high-burn growth startup, the CFO persona should bias toward growth investments above a certain runway threshold. If you're bootstrapped, it should bias toward conservation. The file is yours; the bias is yours.

Voice (1 line)

How the CFO talks. Mine: "Precise, numbers-first, short. Doesn't soften bad news."

Biases (4 lines)

The CFO's preferences:

• Pull toward longer runway over more ambitious campaigns
• Pull toward quarterly billing over annual commits
• Pull toward measurable ROI windows over "strategic" framing
• Pull toward override-with-log over decision-by-vibe

Refusals (5 lines)

This is the heart of the file. What the CFO won't do:

• Spend approval that drops runway below 12 months without written override
• Pricing recommendations without CRO + CEO collaboration
• Investor updates without 13-week forecast attached
• Vendor evaluation without alternative pricing checked

Operating rules (6 lines)

The behavioral defaults:

• Always quote runway in months, not years
• Always show 3 scenarios (best, base, worst), not single number
• Always name the assumption behind a recommendation
• Always log overrides
• Never recommend "the safe option" — recommend the option with the best risk-adjusted return given current runway

Composition (3 lines)

Who CFO reads from and writes to:

• Reads from: metrics-bot (Layer 2 data), CEO Staff brief
• Writes to: CEO scenarios, CRO pricing decisions, EA spend approval workflow

Skills roster (40 lines)

The 10 numbered skills with one-line descriptions, each linking to its own SOP file at c-suite/cfo/skills/<skill>.md.

That's 72 lines total. The whole CFO sits on disk in a folder smaller than most JSON config files.

Installation — 5 Minutes End To End

The repo is at github.com/plugtheliam/myinc-os. MIT licensed.

Step 1 — Clone the repo (30 seconds)

git clone https://github.com/plugtheliam/myinc-os.git
cd myinc-os

Step 2 — Copy the CFO into your agent's skills folder (1 minute)

If you're on Claude Code:

mkdir -p ~/.claude/skills/cfo
cp c-suite/cfo/SKILL.md ~/.claude/skills/cfo/
cp -r c-suite/cfo/skills ~/.claude/skills/cfo/

If you're on Codex / Cursor / Gemini CLI, copy AGENTS.md instead of SKILL.md and drop it at your project root:

cp c-suite/cfo/AGENTS.md ./AGENTS.md

Same content, different wrapper. See docs/CLAUDE-VS-CODEX.md in the repo for the migration mapping.

Step 3 — Customize the \[bracketed\] placeholders (2 minutes)

Open ~/.claude/skills/cfo/SKILL.md in your editor and replace:

• \[Founder Name\] → your name
• \[Company Stage\] → seed / Series A / bootstrapped / revenue / etc.
• \[Policy: Runway Threshold\] → your runway threshold (default 12 months)
• \[Currency\] → USD / EUR / KRW / etc.
• \[Reporting Cadence\] → monthly / quarterly

These are the variables the file uses to scope its refusals. The default 12-month threshold works for most early-stage solo founders. If you're profitable, set it to 6 months. If you're pre-revenue with VC money, keep 12.

Step 4 — Test the install (1.5 minutes)

In Claude Code (or your equivalent), trigger:

have the CFO run /pre-spend-check on this vendor: \[paste vendor info + amount + commitment\]

You should see the structured output I showed above. If the output is generic, the SKILL.md wasn't loaded — check that the file is at the right path and the YAML frontmatter is intact.

That's the whole install. You now have a CFO file that will refuse your bad spending decisions for the next 12 months, until you edit it.

What Happens In Week 1

Here's the actual rhythm I followed in my first 7 days running the CFO file. Your mileage will vary, but the shape is repeatable.

Day 1 (install day) — Test it on one pending decision. The first refusal is uncomfortable; your instinct is to override it. Don't. Sit with it for an hour. See if the alternatives the file proposed are actually viable. They usually are.

Day 2-3 — Add the /pre-spend-check trigger to your manual spend approval workflow. For me, this is a 30-second pre-flight check before signing anything over a meaningful monthly commitment. The file becomes a reflex.

Day 4-5 — Run /runway-snapshot once. Then again 24 hours later. Compare the two. The file should produce a consistent forecast — if it doesn't, your input data is moving (which is interesting in itself; treat it as a signal).

Day 6 — Run /vendor-contract-review on your three biggest recurring spends. The file will identify negotiation levers you didn't consider. I found two contracts I could move from annual to quarterly without the vendor pushing back.

Day 7 — End-of-week retro. Check the .myinc/overrides.jsonl log. How many overrides did you write this week? My first week: 1 override (justified). My current rate: 1 every 3-4 weeks. The compounding effect is real.

Outcomes I observed in week 1:

• Caught one pending renewal that would have wasted recurring cost on a tool I wasn't using
• Refused one paid ads top-up that would have dropped runway below threshold (re-allocated to organic the same week)
• Drafted an investor update in 30 minutes instead of 3 hours
• Started referencing the runway snapshot in CRO and CMO decisions automatically

The compound effect isn't the time saved per skill. It's that the file thinks the thoughts I would have skipped. The breakeven analysis I would have eyeballed. The override path I would have rationalized. The alternative I would have remembered in the shower.

Why This Beats Asking A Single AI Chat The Same Question

You could, in theory, paste your runway numbers into any AI chat and ask for a pricing analysis. People do this. It mostly doesn't work, and here's why:

A single chat tries to be CFO, CMO, CTO, and CEO at the same time. It doesn't know which mode to enter. It defaults to "helpful" — which means agreeable. Which means it will recommend the vendor contract you wanted approved before you started typing. It is a yes-machine. Yes-machines that agree with everything you say are worse than no advisor.

A file is different. The persona is fixed. The biases are explicit. The refusals are listed. When you trigger /pre-spend-check, the file knows it's the CFO. It knows it refuses spend below 12 months runway. It does not pivot into a "well, here are some perspectives to consider" mush. It says no, with reasons.

The other thing files give you: edit-once, fix-forever behavior. When the CFO file makes a mistake (and it will, in your first week), you don't fix the output. You fix the file. The fix compounds — every future invocation inherits it. After three weeks of edits, your CFO file is no longer a generic CFO. It is yours, tuned to your business, your runway, your refusal thresholds.

You cannot get this from a chat.

What Else Is In The Repo

The CFO is 1 of 11 advisors I open-sourced this week. The full pack:

• CEO Staff — pulls signals from every other advisor, packages priority briefs, drives standup-sync
• EA — protects time, triages inbox, prepares meeting briefs, kills calendar conflicts
• CFO — this one (cash, pricing, runway, vendor)
• CEO — strategic synthesis, vision check, scenario war-rooms
• COO — system builder, vendor evaluator, incident reviewer
• CSO — competitive intel, positioning, moat thesis
• CPO — roadmap owner, user research synthesis, feature prioritization
• CTO — architecture, security, infrastructure, build-vs-buy
• CMO — distribution, brand, content strategy, launch coordination
• CRO — pipeline, deals, partnerships, revenue forecast
• Advisor — devil's advocate, blind spots, pre-mortems, dissent

Each is the same shape: persona + voice + biases + refusals + operating rules + composition + 10 numbered skill SOPs. 110 paste-ready files total. MIT licensed. Both Claude Code (SKILL.md) and Codex / Cursor / Gemini CLI (AGENTS.md) formats from day one.

I covered the architecture (11 advisors, two-layer system, why refusals matter) in a separate piece yesterday. This piece is just about installing one and using it well.

The 60-Minute Move (If You're Picking One This Weekend)

Don't try to install all 11 in one sitting. Pick one. Spend 60 minutes on it Saturday. That's enough to get the value.

If you're picking the CFO specifically:

0:00–0:05 Clone the repo + copy SKILL.md to your agent's skills folder
0:05–0:15 Customize the \[bracketed\] placeholders + set your runway threshold
0:15–0:30 Test /pre-spend-check on a pending decision; observe the output
0:30–0:45 Test /runway-snapshot with your current financials
0:45–0:60 Edit the SKILL.md to match your actual operating style (tone, thresholds, what to flag)

By Sunday: working CFO file. Test it on your next 5 spending decisions. Within 30 days, it's tuned. Within 90 days, you're not making the same money mistakes you were making 3 months ago.

Where To Start

The repo: github.com/plugtheliam/myinc-os/tree/master/c-suite/cfo

If you install it, push back, or improve a skill: PRs welcome. Industry-specific variants (DTC CFO, SaaS CFO, agency CFO) and translations both go on the roadmap.

If you're a solo operator running anything with recurring spend — and you are — the CFO file is the highest-leverage first install you can make this weekend.

Three steps. 72 lines. 5 minutes. The file does the thinking you would have skipped.

— Liam