# 09. Minimum Tool Set

A first project rarely stops because it lacks a twentieth service. It stops because source files live in one place, the prompt in another, the result has no name, and a new subscription has already been bought “just in case.” This lesson helps you build a short working route, not a logo collection.

A **tool set** is the smallest collection of places and programs where every needed job has a home: you plan, store material, prepare an image, create motion, assemble the result, and preserve a lesson. One tool can cover several roles; unused roles can stay out.

```learner-callout
{"stable_id":"c100-key-point-job-before-tool","role":"key-point","title":"Key point","body":"Name the job and needed result first. Choose a tool class second. A service name, price, availability, and feature set are not permanent properties of your process."}
```

## 1. Start with a job, not a brand

Imagine a small first project: you have an accepted still of an object and need a short moving fragment for later assembly. Do not search for “the best AI.” Record the input and output first: an accepted still goes in, a short clip comes out, and both files plus a test note need a place to live.

That already gives you a route: planning → storage → image or control frame → video → editing → lesson note. A visual board and separate quality improvement appear only when the task really needs them.

```learner-callout
{"stable_id":"c100-observation-roles-not-products","role":"observation","title":"Observation","body":"A role is a job in the project, not an app name. One service may cover two roles, while a good first project may need neither a board nor separate quality improvement."}
```

## 2. Seven possible working roles

1. **Planning** — defines the task, script, checks, and working text.
2. **Storage** — keeps source files, versions, and accepted results findable.
3. **Visual board** — lets you see references, prompts, variants, and failures together when comparisons are numerous.
4. **Images** — creates or preserves the appearance of a character, object, location, or control frame.
5. **Video** — creates or changes one moving fragment from clear inputs and a feasible change.
6. **Editing** — joins accepted fragments, exact text, sound, graphics, and pauses into the main file.
7. **Quality improvement** — polishes a strong selected result; it does not repair a broken face, action, or composition.

Image-to-video, video-to-video, and editing an existing clip are different job classes. Do not compare them merely because every product page says AI.

## 3. Build the first minimum

Complete only the lines your project needs. This is a copyable card, not a required shopping list.

```template
Project task:
Where I define the task and working text:
Where I store source files and versions:
Where I prepare or check images:
Where I create or change video:
Where I assemble the final file:
Where I preserve a lesson after the test:
Optional role I am not taking now:
```

For the teaching example, one clear planning note, one local versioned folder, one selected image or video route, a program that assembles received files, and a short Markdown note beside the project are enough. This is not a “poor stack.” It is a stack where you can find the cause and the next action.

```learner-callout
{"stable_id":"c100-technique-small-first-stack","role":"technique","title":"Technique: one route per role","body":"Give every needed job one clear place. Add a tool only when it closes a specific uncovered job or replaces one existing step after comparison."}
```

## 4. Choose a tool through a short test

Before paying, trust neither someone else’s ranking nor an old showcase. Open the official page and current interface, then compare no more than three candidates on the same small piece of material.

```template
Working role:
Exact input:
Required output:
Mandatory features:
What I need to check on the official page and in the interface:
What I will test on my own small material:
Rights, terms, and limits important to the project:
Fallback:
Decision after the test: assign role / do not use / check one more candidate
```

Check the required route and input file types, format and export, regional availability, system requirements for local software, result-use rules, and current limits, credits, and price. The last four age especially quickly.

```learner-callout
{"stable_id":"c100-caution-volatile-tool-facts","role":"caution","title":"Caution","body":"Price, limits, availability, rules, and interface controls change. A historical table or review can suggest questions, but it cannot prove that a feature is available to you today."}
```

---

## 5. Folder, board, and agent: three different helpers

A folder is like a cabinet: clear names and versions keep order. A board is like a working wall: it makes reference, working text, variants, and corrections easy to compare. Choose a board when you need to see many visual comparisons; choose folders when the material is small and strict structure helps. You may combine them, but neither is a ritual.

An agent helps when you need to hold written project rules, connect a script to frames and references, compare variants by one criterion, or preserve a lesson from a test. It should not quietly change an accepted story or present a guess about a service as fact. Give it a source and an exact task; keep creative alternatives distinct from confirmed information.

## 6. How to read dated examples

This reference may contain examples of official places to check agent work, boards, generation, quality improvement, and editing. They orient you to a job class; they do not tell you to buy a particular product.

```learner-callout
{"stable_id":"c100-source-dated-reference","role":"source","title":"Dated reference","body":"For a model or interface, the check date and evidence status matter. An official capability, availability in your account, and a result on your material are three different checks."}
```

When a particular route, limit, or compatibility point matters, find a current model adapter or official page with a check date. Turn an old example into a question for a fresh test, not a rule.

## 7. Practice: assign places to one project

Take one real project and complete the two cards above. Leave optional roles empty. Then test one chosen candidate on a small piece of material—or honestly mark that the test has not happened yet.

```learner-callout
{"stable_id":"c100-practice-one-project-stack","role":"practice","title":"Practice","body":"Name one place for every needed job, one fallback, and one small test. Do not buy the full set before at least one tool has proved its role on your material."}
```

```learner-callout
{"stable_id":"c100-checkpoint-stack-ready","role":"checkpoint","title":"Checkpoint","body":"The lesson is complete when every needed job has a clear place, unnecessary roles were not added, volatile facts are marked for fresh verification, and source files, results, and lessons will not disappear between steps."}
```

Next: once the minimum route is assigned, use it for one short project. Do not turn this toolbox into a mandatory linear stage or a prompt library.