ダウンロードフォルダの整理: how to decide what you need

Most advice about ダウンロードフォルダの整理 erabikata skips the only step that changes the outcome, which is looking at what is actually in the folder. Recommendations get made on vibes: the folder feels bad, therefore automate it. Then the setup runs for three weeks and gets switched off, because it was solving a problem the reader did not have. The method below takes about fifteen minutes and produces a decision you can defend, rather than a tool you have to justify.

Measure the folder before choosing anything

Four numbers are enough, and every one of them can be read from a terminal window in seconds.

The first is the total count. Whether the folder holds three hundred items or eight thousand decides whether the backlog can be cleared by hand at all. Anything under roughly five hundred is an afternoon of sorting, which means the real decision is only about stopping it from happening again.

The second is the age distribution. Count how many items have not been touched in ninety days, then in a year. If eighty percent of the folder is older than ninety days, the folder is not an inbox any more. It is an archive with a misleading name, and archiving it wholesale is safe, fast, and free.

The third is the shape of the contents. A quick histogram of extensions tells you whether this is one problem or several. A folder that is ninety percent installer disk images and archive files has a deletion problem. A folder full of PDFs and spreadsheets from clients has a filing problem. These need different answers, and the histogram is the fastest way to tell them apart.

The fourth is the arrival rate. Count items added in the last seven days and divide. Ten a day and thirty a day feel identical when you look at the folder, and they are not the same decision at all.

There is a fifth signal worth checking when the histogram is ambiguous, which is where the files came from. macOS records the source of a downloaded file in an extended attribute, readable with xattr, and running that across a sample of the folder shows whether the volume is coming from one place or twenty. A folder fed almost entirely by one web application can often be fixed at the source, by changing that application's download destination, which removes the problem instead of managing it.

ls -1 ~/Downloads | wc -l
find ~/Downloads -maxdepth 1 -mtime +90 | wc -l
ls -1 ~/Downloads | sed 's/.*\.//' | sort | uniq -c | sort -rn | head
find ~/Downloads -maxdepth 1 -mtime -7 | wc -l

Write the four numbers down. They are the only evidence available, and every choice after this point should be traceable to one of them.

Three failure modes, and only one of them is yours

With the numbers in hand, the folder is failing in one of three ways. They look the same from the outside and they need opposite responses, which is why generic advice tends to miss.

Arrivals outpace attention

High arrival rate, low age. Things land faster than anything gets dealt with, and the folder is genuinely current. This is the case where automation is the right answer, because the volume is real and recurring. It is also the least common of the three.

Things cannot be found later

Moderate count, wide spread of ages, and the pain shows up as searching rather than scrolling. The folder is not too large, it is undifferentiated. Automation makes this worse before it makes it better, because moving files into a structure you have not settled on means searching two places instead of one. The fix here is a naming convention and a saved search, not a rules engine.

Nothing can be thrown away

The dominant case. High count, mostly old, and every attempt at clearing stalls on the same twenty files that might matter. The blocking resource is not time or tooling, it is the willingness to make an irreversible call. Any tool that requires you to define what is safe to delete has already lost, because defining that is the thing you cannot do.

The response to the third case is to stop requiring a decision. A dated archive folder that everything older than sixty days moves into preserves every file, empties the working folder, and asks nothing of you. It is the cheapest answer available and it fits the most common failure mode, which is why it is worth ruling in or out before looking at any software at all.

The criteria that actually change the answer

Feature lists are long and mostly irrelevant. Five properties determine whether a given approach will still be running in six months.

Criterion	Why it decides	How to check before committing
Runs without you	A setup that needs launching is a habit, and habits lapse	Does it install a background agent, or is it a command you type
Reversible actions	Determines how cautious you have to be with every rule	Look for an explicit undo, and for a list of which actions it covers
Permission footprint	Some capabilities require full disk access, some need nothing	Read which features trigger a system permission prompt
Survives a new Mac	Configuration you cannot export is configuration you will rebuild	Is there an export file, and is it a readable format
Cost shape	A one-time price and a recurring price fail differently	One-time with paid upgrades, subscription, or free with your time

The permission row is worth dwelling on, because it is the one that gets clicked past. Features that read or delete from the Trash generally need full disk access, granted in System Settings, which is a broad grant to make for a narrow capability. Sorting files within your home folder does not need it. Knowing which half of a tool's feature list you actually want tells you whether that grant is necessary, and the honest answer is often that it is not.

The last row deserves its own section, because it is where most reasoning goes wrong.

Reversibility is the criterion people skip

Every approach can be described by how expensive its worst mistake is, and that number is knowable in advance.

A rules app that offers an explicit undo is the safe end. Hazel, as the common reference point, restores moves, renames, sorting into subfolders, tag changes, and comments through a Revert command, and does not restore copies, uploads, imports, or script actions. That distinction is published, which means the risk can be evaluated before installing anything.

A shell script sits at the other end. It has no preview and no undo. A pattern that matches more than intended moves everything that matched, and the only recovery is a backup. This is not an argument against scripts. It is an argument for a specific habit: run the script with the destructive command replaced by a print statement, read the list it produces, and only then let it act. Doing that once is worth more than any feature comparison.

Somewhere in the middle is doing nothing automatic at all, which has no failure mode except the folder continuing to grow. That is a legitimate choice when the measurements say the folder is an archive rather than an inbox, and it is the option that gets dismissed too quickly because it does not feel like a decision.

Match the caution to the age of the file. Recent items are cheap to get wrong because you remember them. A rule that only touches things older than sixty days is safer than one that fires on arrival, not because the code is better but because everything it touches has already stopped mattering.

Cost shape matters more than cost

Two prices that look similar over a year behave very differently over five.

A one-time licence with paid upgrades at major versions means the tool keeps working after you stop paying, and the next charge arrives when the developer ships a major release, not on a calendar. Hazel is priced this way, at $42 for a single licence and $20 to move up from an earlier version. A subscription means the capability stops when the payments stop, which matters for something that has been silently filing your documents for two years.

The free options have a cost too, and it is paid in a currency that is easy to under-count. A script takes an evening to write and an hour every time macOS changes something underneath it. Folder Actions take longer than that, because AppleScript is unfamiliar to most people who are otherwise comfortable in a terminal. Neither is expensive, but neither is free either, and the comparison is dishonest if the hours are left out.

The useful test is what happens if you stop paying attention for a year. A licence keeps running. A subscription keeps billing. A script keeps running until something changes, and then fails quietly, which is the worst of the three because you find out by noticing that the folder has four thousand items in it again.

Run the decision for two weeks on one folder

Whatever the criteria point at, do not deploy it everywhere. Pick the single folder the measurements identified and run the choice there for two weeks.

Trials are set up for exactly this. An unregistered copy of Hazel runs with full features for fourteen days before dropping back to a limited mode that watches one folder and allows two active rules. That limit is not a restriction during an evaluation, it is a useful shape: one folder and two rules is exactly the size of test that produces a readable result. A free approach deserves the same discipline, with one launch agent and one script.

At the end of two weeks, take the four measurements again. The count, the ninety-day figure, the extension histogram, and the arrival rate. If the count has not fallen, the approach is not working, regardless of how satisfying the setup was. If the count has fallen but finding things has become harder, the destination structure is wrong and the automation is fine. Those two outcomes have different fixes, and only the second measurement round distinguishes them.

Keep whatever you removed. An archive folder that is never opened costs nothing but disk space, and knowing it exists is what makes aggressive sweeping possible in the first place.

One more thing is worth recording during the trial, and it takes no effort: note every time you have to intervene. A file the rule put somewhere wrong, a download you moved by hand because waiting for the sweep was slower, a condition you edited because it caught too much. Three interventions in two weeks is a setup that is basically working and needs a small correction. Ten is a setup that has replaced one chore with a different one, which is the outcome that quietly wastes the most time, because it looks like progress from the outside.

What to change first

Run the four counts before reading another comparison. If most of the folder is older than ninety days, move that portion into a dated archive today and decide about tooling afterwards, against a folder that reflects your actual working habits. When a real comparison is warranted, check what each option does about the step after sorting, because a file arriving in the right folder still leaves you switching between a folder window and a terminal, which is what a file manager with a built-in terminal is aimed at and what a side-by-side comparison tends to leave out. Atriens lists what that costs.