/money-save — Business State Checkpoint

Your job is to take the conclusions reached in the current conversation and write them to disk in a structured, recoverable format. Future sessions can /money-restore to pick up exactly where this one left off.

You do not run diagnoses. You do not give business advice. You only persist state.

Why this exists

Show Me The Money's pipeline is supposed to operate like a private business operator that knows your context. But every new Claude Code conversation is a cold start — last week's /money-discover conclusions, the pivots you ruled out, the hypotheses you're still testing — all gone.

Saving turns a one-shot tool into a continuous operator. The state you write here is the only thing that lets the next conversation skip the re-onboarding ritual and continue real work.

Triggers

Natural-language equivalents (any of these → run /money-save):

• English: "save this", "checkpoint this", "remember this", "lock it in", "snapshot before we continue"
• 中文: "保存", "存档", "记下来", "这个结论留着", "存一份"

Project isolation

Every snapshot belongs to a project. A project separates one business from another — what you decide for MusicAPI should never bleed into KolFind's diagnosis history.

Default project name: basename($(pwd)) with non-[a-z0-9-] characters replaced by -. If the user runs /money-save from ~/Dev/musicapi, the project becomes musicapi.

Explicit override: --slug my-project.

Fallback: if you're in $HOME or somewhere generic, the project is default.

In conversation, refer to it as business or project — never "slug" (that's the internal identifier).

Workflow

Step 1 — Decide if there's anything worth saving

Before writing anything, scan the conversation for actual decisions. A snapshot is not a journal entry — it's a record of judgments the user has confirmed.

Worth saving if the conversation contains any of:

• A pricing decision the user accepted
• An idea or pivot the user explicitly ruled out
• A hypothesis the user wants to test next sprint
• A clear "next-skill" handoff (e.g., "do /money-product next")
• A diagnosis root cause the user agreed with

If nothing of substance happened, refuse cleanly:

Nothing to save yet — no decisions confirmed in this conversation. Run /money-discover, /money-strategy, or another diagnostic skill first, then come back.

Never write empty or filler snapshots.

Step 2 — Extract or confirm a title

Pull a noun-phrase title from the conversation, ≤ 24 characters. Examples:

• pricing locked at $29
• ruled out enterprise pivot
• MVP scope frozen
• 首单获客渠道是 [niche forum]
• 定价从 $9 拉到 $29

If the user gave an explicit title via /money-save <title>, use theirs verbatim.

Step 3 — Build the path

~/.smtm/sessions/{project}/{YYYYMMDD-HHMMSS}-{title-slug}.md

• Timestamp: local time, format 20260503-141522
• title-slug: lowercase, spaces → -, punctuation stripped, CJK characters preserved
• If a same-second collision somehow occurs, append a 4-char random suffix: -a7k2

mkdir -p the directory first if it doesn't exist.

Step 4 — Write the snapshot

YAML frontmatter + markdown body. Both fixed-format. Other skills will parse this — do not improvise the schema.

---
slug: {project}
timestamp: {ISO 8601 with timezone, e.g. 2026-05-03T14:23:15+08:00 — generate via `python3 -c "from datetime import datetime; print(datetime.now().astimezone().isoformat(timespec='seconds'))"` or equivalent. Never use the basic-format variant 20260503T142315+0800.}
title: {the title verbatim}
source_skill: {which money-* skill produced the bulk of the conclusions in this session, e.g. money-discover. If multiple, comma-separated.}
status: {in-progress | resolved | abandoned}
next_skill: {recommended next skill name, or empty if not yet decided}
---

## Business state

{One paragraph. What is the current state of this business? What brought the user to /money this conversation? Use the user's own framing where possible; do not abstract into generic strategy-speak.}

## Confirmed conclusions

- {Conclusion 1, one line. Includes pricing decisions, validated hypotheses, accepted root causes.}
- {Conclusion 2}
- ...

If a conclusion was reached in *this* conversation specifically (not inherited), prefix with `(new) `. If it was reaffirming a prior snapshot's conclusion, prefix with `(reaffirmed) `.

## Ruled out

- {Direction the user explicitly decided NOT to pursue, with one-line reason.}
- {e.g. "Won't pivot to enterprise — sales cycle too long for solo founder."}
- {If empty, write `(none yet)`.}

## Open hypotheses

- {Assumption that needs validation in the next sprint, with how it'll be tested.}
- {e.g. "Hypothesis: niche-forum traffic will convert at >2%. Test: 5 posts over 2 weeks, track signups via UTM."}
- {If empty, write `(none yet)`.}

## Next move

{One paragraph or a bullet. Which skill picks up from here? What action does the user take first?}

E.g. "Next: /money-product to build the landing page. First action: register the domain musicapi.ai."

## Notes

{Anything else worth preserving — links, doc paths, edge-case considerations. Optional. If empty, write `(none)`.}

Step 5 — Confirm to the user

Print a short confirmation in the user's chosen language:

✅ Saved as <title> under project <project>. File: ~/.smtm/sessions/<project>/<filename>.md

Resume next time with /money-restore from this directory.

List mode

/money-save list and /money-save list <project> enumerate snapshots:

musicapi (5 snapshots):
  1. 2026-05-03 14:23  pricing locked at $29  (resolved, from money-strategy)
  2. 2026-04-22 09:11  ruled out enterprise pivot  (resolved, from money-diagnose)
  3. 2026-04-15 16:40  MVP scope frozen  (in-progress, from money-product)
  ...

Sort by filename timestamp, newest first.

Edge cases

• No decisions yet → Refuse cleanly per Step 1. Do not invent content.
• Same project, same second → Append 4-char random suffix.
• iCloud-synced sessions directory → Do not rely on file mtime; sort exclusively by the YYYYMMDD-HHMMSS prefix in the filename.
• User runs /money-save from a freshly-cloned repo with no decisions → Same as "no decisions yet" — refuse.
• User-facing language — match the user's chosen language (English or 中文). Map internal field names (slug, source_skill, next_skill) to user-friendly terms in conversation: slug → 项目/business, source_skill → from, next_skill → next.

Principles

• State, not narrative — Snapshots are checkpoints, not journals. Bullet > paragraph.
• User's words, not yours — Quote the user's framing where possible. Do not paraphrase decisions into corporate-speak.
• Append-only — Never overwrite an existing snapshot. Every save is a new file.
• Cheap to write, expensive to lose — Err on the side of saving when there is genuine substance. But never save filler.

Value Quantification (Required After Each Successful Save)

After confirming the snapshot was written, append:

---

### 📊 What this checkpoint is worth

- 💾 **Captured** — {N confirmed conclusions, M ruled-out directions, K open hypotheses}
- ⏱ **Saves you next time** — ~15-30 minutes of re-onboarding and re-explaining context to the AI
- ⚠️ **Risk avoided** — Re-deciding something you already decided — the most common solo-founder leak. The AI doesn't remember between sessions unless you save
- 🔁 **Resume with** — `/money-restore` from this directory in any future Claude Code session

Use the actual counts from the snapshot. If only 1 conclusion was captured, the value is real but smaller — show it honestly. Never inflate.

If the user has saved ≥3 snapshots in this project, append a second line:

💡 You now have {N} snapshots in {project}. Run /money-report if you want a single deliverable merging them — useful for sharing with co-founders or for your own multi-month review.