/money-learn

name: money-learn
description: "Manage project learnings — small, atomic, validated patterns that the agent should remember across all skills and sessions. Different from /money-save (which captures full session state); learnings are individual insights that get auto-loaded into every other money-* skill's context. Use when the user has just discovered something worth remembering — a customer pattern, a pricing insight, a channel that works, a failure mode. Triggered by: 'remember this', 'log a learning', 'this is a pattern', 'show learnings', 'what have we learned', '记住这个', '存入经验', '查看经验库'."

/money-learn — Project Learnings Manager

Your job is to maintain a project's `learnings.jsonl` — a JSONL file of validated patterns that other skills auto-load when they run. Each learning is one row, atomic, citable, and worth remembering across all future sessions.

**Learnings are NOT snapshots.** A snapshot captures full session state. A learning is a single, durable insight that should influence future thinking even when no specific snapshot is being restored.

---

Why this exists separately from /money-save

| | `/money-save` | `/money-learn` | |---|---|---| | Granularity | Full session state | One pattern per row | | Frequency | After a major decision | Whenever a pattern is observed | | Auto-loaded? | Only when `/money-restore` is called | **Yes** — every money-* skill loads recent learnings | | Mutability | Append-only snapshots | Add, search, prune, supersede | | Use case | "Resume from this state" | "Remember this pattern always" |

A founder discovers things like:

• "Cold email open rates are 4x higher when the subject is a specific revenue number, not a benefit promise."
• "Our ICP doesn't read X/Twitter — they live in Reddit r/SaaS."
• "Pricing at $39 converts 30% better than $29 even though it's higher."
• "Customers who upgrade past the $99 tier always cite the team-seat feature."

These are atomic patterns. Each gets one row in `learnings.jsonl`. They're auto-injected into every future `/money-discover`, `/money-strategy`, `/money-content`, etc., so the agent stops re-suggesting things you've already invalidated.

---

Triggers

| Command | Behavior | |---|---| | `/money-learn` | Show recent 5 learnings for current project | | `/money-learn add` | Interactive: extract a learning from current conversation | | `/money-learn add "<one-line pattern>"` | Add a learning with explicit text | | `/money-learn search <query>` | Search learnings by keyword/topic | | `/money-learn list` | List all learnings for current project | | `/money-learn list <project>` | List learnings for another project | | `/money-learn prune` | Interactive: review old/contradicted learnings, mark as superseded or remove | | `/money-learn export` | Output all learnings as a markdown table | | `/money-learn promote <L-id>` | Promote a project-local learning to the portfolio layer (see below) | | `/money-learn portfolio` | Show portfolio-wide learnings shared across every project | | `/money-learn portfolio search <query>` | Search portfolio-wide learnings | | `/money-learn portfolio demote <L-id>` | Move a portfolio learning back to a single project (if it turned out to be context-specific) |

**Natural-language equivalents**:

• "Remember this", "Log this learning", "This is a pattern worth keeping"
• "What have we learned", "Show learnings", "Show me the learnings"
• "记住这个", "存入经验", "这是一个模式", "查看经验库"

---

Schema

Each line in `~/.smtm/projects/{slug}/learnings.jsonl` is one JSON object with this fixed schema:

{
  "id": "L-{4 hex chars}",
  "captured_at": "ISO 8601 with timezone",
  "from_skill": "name of the skill that generated this learning, or 'manual'",
  "category": "one of: pricing | channel | icp | positioning | conversion | retention | ops | tech | competition | personal",
  "pattern": "One sentence stating the pattern. Imperative or declarative; no hedging.",
  "evidence": "Concrete evidence supporting the pattern. Specific numbers, dates, quotes preferred.",
  "confidence": "validated | emerging | hypothesis",
  "supersedes": "id of an older learning this replaces (or null)",
  "tags": ["arbitrary", "free-form", "tags"]
}

Confidence levels

• **validated** — At least 2 independent observations or 1 quantitative result with N>30. Acts on this freely.
• **emerging** — One strong observation, not yet replicated. Other skills consider it but don't lock in.
• **hypothesis** — Pattern noticed once, untested. Surfaced for awareness only.

Categories (closed list)

`pricing`, `channel`, `icp`, `positioning`, `conversion`, `retention`, `ops`, `tech`, `competition`, `personal`

If a learning doesn't fit any category, force a fit — usually it's `personal` (about the founder) or `ops`. Avoid creating new categories; the closed list keeps the auto-load logic predictable.

---

Workflow

`/money-learn add` (interactive mode)

Walk through a 5-step extraction:

1. **What pattern?** — One sentence. If the user gives a paragraph, paraphrase to one declarative sentence. 2. **What's the evidence?** — Specific. "5 customers said X" not "customers say X". If evidence is vague, reduce confidence to `hypothesis`. 3. **What category?** — Pick from the closed list. 4. **Confidence?** — Default to `emerging` unless evidence is N≥30 or 2+ independent observations. 5. **Does this supersede an older learning?** — Search for similar patterns, ask the user.

Then write the JSON line to disk and confirm. Print the row that was added.

Auto-extraction from conversation

If the user invokes `/money-learn` without arguments and there's a clear pattern in the recent conversation (e.g., they just said "wow, the $39 price converts way better than $29"), auto-propose the extraction:

> I noticed a pattern in this conversation. Want to log: > - Pattern: "Pricing at $39 converts 30% better than $29 in our ICP" > - Evidence: "{quoted observation from conversation}" > - Category: pricing > - Co