Write a complete OCI image layout under `out` (created if needed): `out/oci-layout`, `out/index.json`, and `out/blobs/sha256/<...>` blobs. `layers` a

Serialize a value with postcard, compress with zstd, hash, and write. This is the standard flow for typed metadata sections (HEADER, DEPS, PROVENANCE

When an identity becomes the active default, automatically activate the server profile bound to it (if any). Looks for a named profile in `~/.atomic/

Add multiple files to tracking in a single write transaction. This is much faster than calling `add()` in a loop because it avoids opening a separate

Add content bytes.

Add an edge to the graph.

Add a bidirectional edge pair through the cached writer, mapping storage errors into the apply error type. This is the single edge-writing entry poin

Add a canonical bidirectional GRAPH edge pair (forward + PARENT reverse), mirroring both rows into INODE_GRAPH when an inode is provided.

Record an error for a path.

Add an error.

Add an error.

Add an insertion to this conflict.

Adds a line with content, tokenizing into leaves.

Add a local-only change hash.

Add a modified file to the result.

Add a moved file to the result.

Add an order conflict.

Add a progress bar to a multi-progress container. # Arguments `mp` - The multi-progress container `total` - The total number of units `message` - Th

Add a recorded file path.

Add a new remote.

Add a new remote and set it as the default. # Arguments `name` - The name for the new remote `url` - The URL of the remote repository

Record a skipped file.

Add a skipped file path.

Adds a token to a line. # Arguments `branch_id` - The line to add the token to `after` - The leaf to insert after (None for start of line) `kind` -

Add a warning message.

Add a warning message.

Returns all affected file paths in a single iterator. Order: modified files first, then added, then deleted. This is the list passed to `RecordOption

List all known predicates.

Allocates a new branch ID.

Allocate a new leaf ID.

Allocate a new node ID This is thread-safe and guaranteed to return unique IDs.

Applies a Delete operation to mark a line as deleted. # Behavior 1. Verifies the branch exists 2. Verifies the branch is not already deleted 3. Upda

Applies an Insert operation without nested leaf operations.

Applies a line change from the analyzer. This is the integration point between the `LineAnalyzer` and the builder.

Applies a Reparent operation to change a branch's chain position without touching its state or content. # Behavior 1. Verifies the branch exists. 2.

Applies a Restore operation to restore a deleted line. # Behavior 1. Verifies the branch exists 2. Verifies the branch is currently deleted 3. Updat

Apply a stash to the current working copy. Reads raw file bytes from the sidecar directory (`.atomic/stashes/<view_name>/`) and writes them to disk.

Get the raw bytes of the identity ID.

Get the raw bytes.

Mark this as a directory.

Open the Atomic repository at `root` and return `(view_name, merkle_base32)` for the current view. Returns `None` if the directory is not an Atomic r

Format an author name in default style. Use this for displaying author names in logs and history. # Arguments `text` - The author name to format #

Denominator for the headline metric: changes authored by a person or an AI. Excludes System and NeedsAttention.

Builds the conflict.

Build an `Author` for an agent turn, using the session's identity info. This is the main entry point called by `record_turn()`. It resolves the agent

Build git metadata for the change's unhashed field.

Build a complete new learnings section (heading + markers + content).

Get the byte-progress bar style. Format: `⠋ [████████░░░░░░░░] 1.5 MB/3.0 MB Message (50%)`

Get the total number of changes.

Iterate over all changed files (added, deleted, modified, moved).

Iterate over only the change operations (non-equal).

Set all covered change hashes at once.

Set all explained change hashes at once.

Check if a change can be unrecorded from a view. This verifies that: 1. The change is in the view 2. No other changes in the view depend on it (unles

Determine whether an Insert hunk is a Prepend, Append, or Middle insert. For multi-vertex files (files that have been recorded at least once with per

Get the effective claude CLI path.

Clean up a directory on error (e.g., failed clone). Silently removes the directory and all its contents. Does nothing if the path doesn't exist. # A

Set the code change statistics.

Collect children of a directory from the graph. Uses the tree traversal module to collect all files and directories under the specified parent path,

Collect sessions stamped with `run_id` from the canonical session store.

Collect all files in a directory, respecting ignore patterns. # Arguments `root` - The root directory to walk `options` - Status options controlling

Collect all files in a directory, respecting ignore patterns and custom rules. This is the full version that accepts optional [`IgnoreRules`] for pat

Format a command or code snippet in bold cyan. Use this for displaying commands the user should run. # Arguments `text` - The command to format #

Convert a `GraphNode<Option<Hash>>` to a [`CompactGraphNode`]. # Errors Returns [`FormatError::HashNotFound`](super::super::error::FormatError::Hash

Compress all chunks in parallel and return compression statistics. This is the same as [`compress_chunks_parallel`] but also returns [`CompressionSta

Compute the full `RemoteDelta` for a given view. This is the main entry point for sync operations. It: 1. Runs the dichotomy algorithm to find the d

Determine the order to insert a set of changes respecting dependencies. This performs a topological sort of the changes based on their dependency rel

Compute the Longest Common Subsequence using dynamic programming. Returns a list of matching positions in both sequences.

Move to the next side of a conflict. # Arguments `id` - The conflict ID (must match the begin call) `changes` - Optional change hashes for this side

Get a reference to the conflicts detected so far.

Get conflicts filtered by type.

Consolidate raw tool nodes into Decision nodes. Scans the graph for recognizable sequences of unclassified tool nodes and replaces each sequence with

Whether `change_id`'s spans have already been loaded.

Read only CONTENT sections, skipping everything else. This reads the raw file content chunks without loading graph or semantic operations. Useful whe

Compute the content hash of this tag. Covers identity + position + annotation fields. The `metadata` field is excluded (same pattern as `Change.unhas