atl decided

Search every surface that can hold a decision, and report honestly when there is nothing there.

atl decided "brief and stop"

The question it answers

Before you write a rationale — why the behaviour is this way, what it deliberately is not — there is a question worth asking: does the record actually hold the decision you are about to assert?

It is easy to skip, because an authored justification reads as authoritative and nothing about it looks wrong. A claim that nobody ever decided survives review exactly as well as one that was.

atl decided is the cheapest possible way to check. One command, no state, no index — a direct search across the decision surfaces.

What it searches

Widest-first, and only the roots that exist:

root	what lives there
.atl/docs	settled decisions
.atl/brain-storms	the discussions that produced them, including rejected options
.atl/wiki · .atl/journal	current truth and history
.claude	installed team knowledge
docs · cli · core · teams	the site, and the code that implements decisions

cli/ is in the list deliberately. A decision is sometimes recorded only in the code that implements it — one real case was settled by a single line of a command's help text and by nothing else.

This is wider than what the per-prompt retrieval hook indexes, on purpose. Retrieval has to be selective; this is a direct search whose whole job is to be exhaustive, and where the expensive outcome is a false negative.

Vendored and generated trees (node_modules, vendor, .git, worktrees) are skipped.

Reading a zero result

0 matches for "brief and stop"
searched: .atl/docs .atl/brain-storms .atl/wiki .claude docs cli core teams

That is a fact about this QUERY, not about the record. A decision written
in other words is indistinguishable from one that does not exist, so try
the synonyms before you draw a conclusion from this.

The empty case is the one this command exists for, which is why it prints the roots it searched instead of nothing at all.

But read the caveat literally. A text search cannot tell you that no decision exists — only that these words are not in these files. A term you did not think of looks exactly like a decision nobody made. Try the other wording before concluding.

The output also names only the roots that were actually present. A directory that does not exist is not searched and not claimed, so the zero result never implies coverage it did not have.

Related

• atl learnings — the write side of the knowledge loop this searches.
• Concepts — how the decision surfaces relate to each other.

The per-prompt ranked surface (atl retrieve) is the selective counterpart: it may return nothing, and it is not exhaustive. This command is the one you reach for deliberately, when a zero result is the answer you need.