.claude/atomic.toml

The repo-scoped config file. It is committed, so it applies to everyone working in the repository, and it holds settings that belong to the project rather than to you.

There are two config files and they do not overlap:

Your personal preferences live in ~/.atomic/config.toml and never reach the repository. Facts about the project live here. A teammate cloning the repo should get the same code-index exclusions you have; they should not inherit your update settings.

atomic repo init creates the file with a scope marker. Everything else you add by hand.

Keys

Key	Type	Default	What it does
scope	string	none	Declares this directory as "repo" or "realm"
[code] ignore	string array	empty	Glob patterns excluded from the code-intel index
[scan] ignore	string array	empty	Glob patterns dropped from the wiki scan entirely
[scan] generated	string array	empty	Globs kept in the tree but skipped for domain content
[repl] idle_timeout	string	1h	How long an idle atomic repl session survives

A complete example:

scope = "repo"

[code]
ignore = ["vendor/**", "*.generated.ts", "dist/**"]

[scan]
ignore = ["fixtures/large-dataset.json", "third_party/**"]
generated = ["*.pb.go", "generated/**"]

[repl]
idle_timeout = "30m"

scope

scope states what a directory is, so nothing has to guess. Without it, a repo root is inferred from git rev-parse --show-toplevel and a realm root is found by looking up your personal <wikis> registry, which means the same directory can resolve differently for two people.

atomic where reports which one answered.

Value	Means
"repo"	A single project. Written by atomic repo init.
"realm"	A root holding several member repos plus a shared wiki. Written by atomic wiki init --scope realm.

Discovery walks upward from the current directory to the filesystem root, taking the first marker of the kind it wants. It crosses .git boundaries deliberately, because a realm root sits above the repos it contains.

A marker naming the other kind is skipped, not treated as an error. atomic doctor warns when scope = "repo" contradicts a realm registered at the same path.

[code] ignore

Patterns here keep files out of the code-intel index. Use it for vendored trees, generated output, and build artifacts, which add nodes and edges without telling you anything about the code you write.

Matching rules, which are not gitignore rules:

Pattern	Matches
vendor/**	contains a slash, so it matches the full repo-relative path
*.generated.ts	no slash, so it matches the basename at any depth
./dist/**	leading ./ is stripped, then treated as a path pattern
build/	trailing slash only, matches nothing — write build/**

There is no negation syntax. An invalid pattern is dropped with a warning rather than failing the run.

Newly ignored files are removed from the index on the next atomic code index or atomic code sync, through the same pruning that handles deleted files.

If your repo gitignores .claude/ wholesale, add a negation so this file is actually committed:

.claude/*
!.claude/atomic.toml

Uncommitted, it only configures your own machine, which defeats the point of a repo-scoped file.

[scan] ignore and [scan] generated

These control what the repo wiki's deterministic scan sees. They are two lists because there are two different things you might want:

Key	Effect	Use for
ignore	the path never appears in the tree at all	vendored dependencies, checked-in fixtures, large data files
generated	the path stays in the tree, marked, and the inferrer skips it for domain content	build output, protobuf output, lockfiles

The scan already starts from tracked files plus untracked files not covered by .gitignore, so anything gitignored is excluded before these run. Reach for [scan] when a committed path should be dropped or flagged.

[scan]
ignore = ["fixtures/large-dataset.json", "third_party/**"]
generated = ["*.pb.go", "generated/**"]

Matching follows the same rules as [code] ignore above: a pattern with a slash matches the full repo-relative path, one without matches the basename at any depth, and ** crosses directories.

Migrating from .signalsignore. These keys replace the repo-root .signalsignore file, whose bare lines are now ignore and whose +-prefixed lines are now generated. atomic update converts an existing file and deletes it; atomic migrate --repo <path> does the same on demand. Until that runs, a repo with no [scan] table still reads .signalsignore, so nothing breaks in the meantime. If both exist, [scan] wins as a whole table — a config declaring only ignore also suppresses the old file's + lines, so the effective rules never depend on a file you forgot was there.

[repl] idle_timeout

How long a named atomic repl interpreter session stays alive with no activity. Accepts a Go duration string: 30m, 2h, 90s.

Resolution order is repo, then user, then the 1h built-in default. Set it here when the project's sessions are expensive to start; set it in ~/.atomic/config.toml when it is your own preference.

State paths

.claude/atomic.toml does not configure where scratchpad bundles, session reports, reminders, or archives live — those resolve per project, not per repo config, under ~/.atomic/<project-key>/. Run atomic where --json to read them; see conventions for the layout.

How the file is read

Loading is lenient by design, because a config problem should never stop you working:

Condition	Result
File absent	Empty config, no error
Unknown key	Warning; the rest of the file still loads
Invalid glob	Pattern dropped with a warning
Malformed TOML	Error, and the caller degrades rather than failing

atomic doctor category 13 reports on this file. An absent file passes, since the file is opt-in. Parse errors, unknown keys, invalid globs, and an invalid scope value all warn. It never fails, so a bad config file cannot block the rest of the check suite.