@orrisai/show-me-the-money 2.2.0 → 2.3.0

package/README.md

@@ -8,7 +8,7 @@
 [![Latest release](https://img.shields.io/github/v/release/iamzifei/show-me-the-money?label=release&color=green)](https://github.com/iamzifei/show-me-the-money/releases)
 [![License: CC BY-NC 4.0](https://img.shields.io/badge/license-CC%20BY--NC%204.0-orange.svg)](LICENSE)
 
-**Current version: `v2.2.0`** · [What's new →](#-whats-new-in-v220)
+**Current version: `v2.3.0`** · [What's new →](#-whats-new-in-v230)
 
 [English](README.md) | [中文](README.zh-CN.md)
 
@@ -51,6 +51,47 @@ Works with **Claude Code**, **Codex CLI**, **Gemini CLI**, and other agents that
 
 ---
 
+## ✨ What's New in v2.3.0
+
+**The founder knowledge base ships with the suite.** Every money-* skill now auto-loads atomic principles distilled from years of solo-SaaS operating notes — battle-tested judgement embedded directly in your AI agent's working context.
+
+### What changed
+
+```
+skills/money/knowledge/atoms/
+  atoms.jsonl                              ← full corpus
+  atoms_solopreneur_psychology.jsonl       ← character, action threshold, focus
+  atoms_market_observation.jsonl           ← shifts, channels, trajectories
+  atoms_agent_infra.jsonl                  ← AI agents, skill design, automation
+  atoms_growth_tactics.jsonl               ← outreach, ads ROI, pricing, conversion
+  atoms_content_meta.jsonl                 ← strategy, AI-content traps, positioning
+```
+
+Each atom is one declarative principle: a market observation, an anti-mainstream take, a hard-won tactic. Distilled from a multi-year archive of working notes — not opinion blog posts.
+
+### Why this matters
+
+Skills used to re-derive the same conclusions every conversation. Now `/money-discover` starts already knowing which market shapes are traps for solo founders. `/money-content` starts already knowing which AI-content patterns kill positioning. `/money-diagnose` cites specific atoms when the user's failure mode matches a known one — and you can trace any recommendation back to its source.
+
+When an atom directly informs a recommendation, the skill cites it by ID:
+
+> "Picking a $29/mo consumer wedge here would hit trap **A-bce2** — agent infra is shifting consumer apps toward UI-less API plays within 12 months."
+
+Click through the source link and you see the original observation, dated, with the original context.
+
+### Two layers of memory now
+
+- **Atoms** (this release) — global, founder-maintained, ship with the package, read-only at runtime. Encode general principles.
+- **Learnings** (v2.2) — project-local, auto-captured per-slug, mutable. Encode this-project-specific patterns.
+
+Together: every skill run starts with both general operating wisdom **and** what we learned about this specific business.
+
+### Maintainer-only: incremental distillation
+
+The pipeline that produces atoms (`scripts/x-distill.mjs`, gitignored) is incremental — only new content gets distilled on subsequent runs, so the corpus grows organically without re-processing the archive.
+
+---
+
 ## ✨ What's New in v2.2.0
 
 **The review panel + cross-session learning.** Eight new skills, plus auto-loading of every prior insight into every future session.

package/VERSION

@@ -1 +1 @@
-2.2.0
+2.3.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@orrisai/show-me-the-money",
-  "version": "2.2.0",
+  "version": "2.3.0",
   "description": "AI agent skills that build and run your business autonomously — from idea to revenue, 24/7. Works with Claude Code, Codex CLI, and Gemini CLI.",
   "keywords": [
     "ai-agent",

package/skills/money/SKILL.md

@@ -275,6 +275,50 @@ Read `~/.smtm/projects/<slug>/skills/` (created by `/money-skillify`). If any cu
 
 Do this once per session, not on every invocation. Track via `~/.smtm/.session-skills-shown-<slug>` touch file (created on show, cleared by `/money-restore` or after 24h).
 
+### Step 5: Auto-load global atoms (founder knowledge base)
+
+Atoms are reusable principles distilled from the maintainer's working notes — battle-tested judgement that should inform every skill run. They live at:
+
+```
+~/.claude/skills/money/knowledge/atoms/
+  atoms.jsonl                              # full corpus
+  atoms_solopreneur_psychology.jsonl
+  atoms_market_observation.jsonl
+  atoms_agent_infra.jsonl
+  atoms_growth_tactics.jsonl
+  atoms_content_meta.jsonl
+```
+
+Per-skill atom slice (load only what's relevant — keep working context lean):
+
+| Skill | Atom categories to load |
+|---|---|
+| `/money-discover` | `market_observation`, `growth_tactics` |
+| `/money-strategy` | `market_observation`, `growth_tactics`, `content_meta` |
+| `/money-content`, `/money-social`, `/money-seo` | `content_meta`, `growth_tactics` |
+| `/money-outreach`, `/money-ads` | `growth_tactics`, `content_meta` |
+| `/money-product`, `/money-quality`, `/money-ops` | `agent_infra` |
+| `/money-finance` | `growth_tactics` (pricing subset only) |
+| `/money-diagnose`, `/money-panel`, `/money-review-*` | ALL (especially `solopreneur_psychology`) |
+| `/money-retro` | ALL |
+| `/money-save`, `/money-restore`, `/money-report`, `/money-learn`, `/money-skillify` | none — state managers don't consume atoms |
+
+Filter to `confidence ∈ {validated, emerging}` by default — skip `hypothesis` unless the user explicitly asks for speculative input.
+
+If matching atoms exist, surface them once at the top of the skill's output:
+
+> 🧠 Loaded N relevant atoms from the founder knowledge base:
+> - A-{id} ({confidence}, {category}): {pattern}
+> - ...
+>
+> These principles will inform the analysis below — citations by `A-{id}` link back to source.
+
+Cite an atom whenever a recommendation is directly informed by it, e.g. *"Picking a $29/mo consumer wedge here would hit the same trap A-bce2 names — agent infra is shifting consumer apps toward UI-less API plays within 12 months."*
+
+If 0 matching atoms (e.g. fresh install, atoms not yet bundled): silently skip. Never fabricate atom IDs.
+
+**Difference from learnings**: atoms are global (founder-maintained, ship with the package, read-only). Learnings (Step 3) are project-local (auto-captured per-slug, mutable). Atoms encode general principles; learnings encode this-project-specific patterns.
+
 ---
 
 ## Auto-Update Check (Once Per Session, /money Router Only)

package/skills/money/knowledge/atoms/README.md

@@ -0,0 +1,67 @@
+# Atoms — Founder Knowledge Base
+
+Atomic, reusable patterns distilled from the maintainer's working notes (X tweets, build logs, retros). Each atom is one declarative principle a money-* skill can cite when reasoning about a user's situation.
+
+## Why atoms?
+
+Skills routinely re-derive the same conclusions every conversation: "API products beat consumer-app GTM for solo founders," "AI-content traps look like growth but kill positioning," "agents replace UI in 12 months." These aren't hot takes — they're principles the maintainer has paid for in time, money, and dead products.
+
+Atoms encode them so every skill run starts with that compounded judgement instead of cold reasoning.
+
+## Files
+
+| File | Contents |
+|---|---|
+| `atoms.jsonl` | Full atom corpus (every accepted atom across every category) |
+| `atoms_solopreneur_psychology.jsonl` | Character, action threshold, focus, mental models |
+| `atoms_market_observation.jsonl` | Market shifts, channel dynamics, industry trajectories |
+| `atoms_agent_infra.jsonl` | AI agents, skill design, tooling, automation infra |
+| `atoms_growth_tactics.jsonl` | Cold outreach, ads ROI, pricing, conversion experiments |
+| `atoms_content_meta.jsonl` | Content strategy, AI-content traps, positioning narrative |
+
+## Atom schema
+
+```json
+{
+  "id": "A-1302",
+  "captured_at": "2026-05-03T04:26:26Z",
+  "source": "https://x.com/<handle>/status/<id>",
+  "source_text": "Original full text",
+  "source_created_at": "2026-05-03T00:35:31Z",
+  "category": "market_observation",
+  "pattern": "One declarative sentence in the source's language.",
+  "confidence": "validated | emerging | hypothesis",
+  "tags": ["1-3", "short", "keywords"],
+  "distill_model": "sonnet",
+  "reason": "1-sentence rationale for atomization."
+}
+```
+
+`confidence`:
+- **validated** — backed by multiple observations or a quantitative outcome
+- **emerging** — one strong observation, not yet replicated
+- **hypothesis** — claim or speculation, untested
+
+## How skills consume atoms
+
+Every money-* skill loads the relevant slice at startup (see Standard Skill Startup in `skills/money/SKILL.md`). The slice rule:
+
+| Skill | Loads |
+|---|---|
+| `/money-discover`, `/money-strategy` | `market_observation` + `growth_tactics` |
+| `/money-content`, `/money-social`, `/money-seo` | `content_meta` + `growth_tactics` |
+| `/money-outreach`, `/money-ads` | `growth_tactics` |
+| `/money-product`, `/money-quality`, `/money-ops` | `agent_infra` |
+| `/money-diagnose`, `/money-panel`, `/money-review-*` | `solopreneur_psychology` + ALL |
+| `/money-finance` | `growth_tactics` (pricing/conversion subset) |
+
+Atoms are read-only at runtime. Skills cite them by `id` when an atom directly informs a recommendation, so the user can trace any conclusion back to its evidence.
+
+## Adding more atoms
+
+The corpus grows in two ways:
+
+1. **X distillation pipeline** (`scripts/x-distill.mjs` — gitignored, maintainer-only). Reads exported tweets and asks Claude to extract atoms category-by-category. Resumable; on subsequent runs only new tweets are processed.
+2. **Manual additions** — append a single line to the relevant `atoms_<category>.jsonl` and the full `atoms.jsonl`. ID format `A-{4 hex chars}`.
+
+Atoms are append-only. To deprecate one, set `confidence: "deprecated"` and add a `superseded_by` field pointing to its replacement — never delete history.