CodexWrapper.ExecFork (CodexWrapper v0.5.2)

Copy Markdown View Source

ExecFork command -- branch an existing session into a new one, non-interactively.

Wraps codex exec fork <SESSION_ID> [PROMPT]. Forking copies the source session's history into a new session and runs the optional prompt there. The source session is left as it was and stays resumable under its original ID.

Usage

config = CodexWrapper.Config.new(working_dir: "/path/to/project")

{:ok, fork} =
  CodexWrapper.ExecFork.new("019a...-source-session-id")
  |> CodexWrapper.ExecFork.prompt("try the other approach")
  |> CodexWrapper.ExecFork.fork(config)

fork.session_id        # the new session
fork.source_session_id # the session that was forked, unchanged

fork/2 is the call to use when the new session ID matters (it usually does). execute/2, execute_json/2, and stream/2 behave the same way as they do on CodexWrapper.ExecResume.

Arguments the CLI accepts

Verified against codex-cli 0.149.0. codex exec fork accepts the codex exec resume flag set without --last and --all, plus --output-schema. It rejects --sandbox, --profile, and --full-auto, so sandbox/2 and full_auto/1 emit the config override -c sandbox_mode="<mode>" the same way ExecResume does.

CLIs without exec fork

An older CLI parses fork as the codex exec prompt and rejects the session ID as an unexpected argument. execute/2, execute_json/2, and fork/2 recognize that failure and return {:error, {:unsupported, :exec_fork}}. supported?/1 checks ahead of time. stream/2 cannot see the exit status or stderr, so when its stream ends without a single line it runs supported?/1 and raises CodexWrapper.UnsupportedError during enumeration if the CLI has no exec fork.

Summary

Types

error()
fork_result()

The outcome of fork/2.

sandbox_mode()
t()

Functions

config(e, kv)

Add a config override (key=value).

dangerously_bypass_approvals_and_sandbox(e)

Bypass all approvals and sandbox. Use with extreme caution.

dangerously_bypass_hook_trust(e)

Run enabled hooks without requiring persisted hook trust. Use with extreme caution.

disable(e, feature)

Disable a feature.

enable(e, feature)

Enable a feature.

ephemeral(e)

Enable ephemeral mode: the new session is not persisted to disk.

execute(exec, config)

Execute the command synchronously, returning a parsed %Result{}.

execute_json(exec, config)

Execute with --json and return the parsed %JsonLineEvent{} list.

fork(exec, config)

Fork the session and return the new session ID alongside the source.

full_auto(e)

Enable full-auto mode.

ignore_rules(e)

Do not load user or project execpolicy .rules files.

ignore_user_config(e)

Do not load $CODEX_HOME/config.toml. Auth still resolves through CODEX_HOME.

image(e, path)

Add an image to attach to the prompt sent after forking.

json(e)

Enable JSON output.

model(e, model)

Set the model.

new(session_id)

Create a fork of the session with the given ID (a UUID or a thread name).

output_last_message(e, path)

Set the output-last-message path.

output_schema(e, path)

Set the path of a JSON Schema file describing the final response shape.

prompt(e, prompt)

Set the prompt to send in the new session after forking.

sandbox(e, mode)

Set the sandbox mode.

skip_git_repo_check(e)

Skip the git repo check.

stream(exec, config)

Execute the command and return a lazy Stream of %JsonLineEvent{}.

strict_config(e)

Error out when config.toml contains fields this Codex version does not recognize.

supported?(config)

Return whether the installed CLI has codex exec fork.

validate(exec_fork)

Check the source session ID before it reaches the CLI.

Types

error()

@type error() ::
  {:invalid_session_id, term()}
  | {:unsupported, :exec_fork}
  | {:exit, non_neg_integer(), CodexWrapper.Result.t()}
  | {:missing_session_id, CodexWrapper.Result.t()}
  | term()

fork_result()

@type fork_result() :: %{
  session_id: String.t(),
  source_session_id: String.t(),
  result: CodexWrapper.Result.t(),
  events: [CodexWrapper.JsonLineEvent.t()]
}

The outcome of fork/2.

session_id is the new session the CLI created; source_session_id is the ID that was forked, exactly as passed to new/1.

sandbox_mode()

@type sandbox_mode() :: :read_only | :workspace_write | :danger_full_access

t()

@type t() :: %CodexWrapper.ExecFork{
  config_overrides: [String.t()],
  dangerously_bypass_approvals_and_sandbox: boolean(),
  dangerously_bypass_hook_trust: boolean(),
  disabled_features: [String.t()],
  enabled_features: [String.t()],
  ephemeral: boolean(),
  full_auto: boolean(),
  ignore_rules: boolean(),
  ignore_user_config: boolean(),
  images: [String.t()],
  json: boolean(),
  model: String.t() | nil,
  output_last_message: String.t() | nil,
  output_schema: String.t() | nil,
  prompt: String.t() | nil,
  sandbox: sandbox_mode() | nil,
  session_id: String.t(),
  skip_git_repo_check: boolean(),
  strict_config: boolean()
}

Functions

config(e, kv)

@spec config(t(), String.t()) :: t()

Add a config override (key=value).

dangerously_bypass_approvals_and_sandbox(e)

@spec dangerously_bypass_approvals_and_sandbox(t()) :: t()

Bypass all approvals and sandbox. Use with extreme caution.

dangerously_bypass_hook_trust(e)

@spec dangerously_bypass_hook_trust(t()) :: t()

Run enabled hooks without requiring persisted hook trust. Use with extreme caution.

Hook trust is what stops a repository from running arbitrary commands the user never approved. Only appropriate for automation that already vets where its hooks come from.

disable(e, feature)

@spec disable(t(), String.t()) :: t()

Disable a feature.

enable(e, feature)

@spec enable(t(), String.t()) :: t()

Enable a feature.

ephemeral(e)

@spec ephemeral(t()) :: t()

Enable ephemeral mode: the new session is not persisted to disk.

execute(exec, config)

@spec execute(t(), CodexWrapper.Config.t()) ::
  {:ok, CodexWrapper.Result.t()} | {:error, error()}

Execute the command synchronously, returning a parsed %Result{}.

A non-zero exit is still {:ok, %Result{success: false}}, as with ExecResume.execute/2, except when the CLI has no exec fork, which returns {:error, {:unsupported, :exec_fork}}. An invalid session ID returns {:error, {:invalid_session_id, value}} without spawning.

execute_json(exec, config)

@spec execute_json(t(), CodexWrapper.Config.t()) ::
  {:ok, [CodexWrapper.JsonLineEvent.t()]} | {:error, error()}

Execute with --json and return the parsed %JsonLineEvent{} list.

fork(exec, config)

@spec fork(t(), CodexWrapper.Config.t()) :: {:ok, fork_result()} | {:error, error()}

Fork the session and return the new session ID alongside the source.

Forces --json. Returns {:ok, fork_result()} when the CLI exits 0 and reports the new thread ID in its thread.started event. The new ID comes from the CLI's output, never from the builder, so source_session_id is always the ID that was forked.

Errors:

  • {:invalid_session_id, value} -- rejected before spawning
  • {:unsupported, :exec_fork} -- the installed CLI has no exec fork
  • {:exit, code, result} -- the CLI exited non-zero (for example, an unknown source session: no rollout found for thread id)
  • {:missing_session_id, result} -- exit 0 without a new thread ID
  • {:timeout, ms} and runner errors, as from execute/2

full_auto(e)

@spec full_auto(t()) :: t()

Enable full-auto mode.

Deprecated upstream, and codex exec fork rejects --full-auto. Emits -c sandbox_mode="workspace-write" instead. An explicit sandbox/2 call is more specific and wins over this.

ignore_rules(e)

@spec ignore_rules(t()) :: t()

Do not load user or project execpolicy .rules files.

ignore_user_config(e)

@spec ignore_user_config(t()) :: t()

Do not load $CODEX_HOME/config.toml. Auth still resolves through CODEX_HOME.

image(e, path)

@spec image(t(), String.t()) :: t()

Add an image to attach to the prompt sent after forking.

json(e)

@spec json(t()) :: t()

Enable JSON output.

model(e, model)

@spec model(t(), String.t()) :: t()

Set the model.

new(session_id)

@spec new(String.t()) :: t()

Create a fork of the session with the given ID (a UUID or a thread name).

The ID is checked by validate/1 when the command runs, not here, so a builder can be assembled from untrusted input and rejected with a tagged error rather than an exception.

output_last_message(e, path)

@spec output_last_message(t(), String.t()) :: t()

Set the output-last-message path.

output_schema(e, path)

@spec output_schema(t(), String.t()) :: t()

Set the path of a JSON Schema file describing the final response shape.

prompt(e, prompt)

@spec prompt(t(), String.t()) :: t()

Set the prompt to send in the new session after forking.

The prompt is passed after a -- option terminator, so a prompt that begins with - or -- reaches the CLI as the prompt rather than as a flag. A prompt that is exactly - still tells the CLI to read the prompt from stdin.

sandbox(e, mode)

@spec sandbox(t(), sandbox_mode()) :: t()

Set the sandbox mode.

Emits -c sandbox_mode="<mode>": codex exec fork rejects --sandbox with unexpected argument, and the config key takes the same three values.

skip_git_repo_check(e)

@spec skip_git_repo_check(t()) :: t()

Skip the git repo check.

stream(exec, config)

@spec stream(t(), CodexWrapper.Config.t()) :: Enumerable.t()

Execute the command and return a lazy Stream of %JsonLineEvent{}.

Forces --json. Raises ArgumentError for an invalid session ID, since a stream has no error tuple to return. The new session ID is in the first thread.started event.

On a CLI without exec fork, enumerating the stream raises CodexWrapper.UnsupportedError with capability: :exec_fork, the stream counterpart of {:error, {:unsupported, :exec_fork}}. Only a stream that ends without output pays for the supported?/1 check; a stream that yields a line, or that the consumer halts early, never runs it. An empty stream from a CLI that does have exec fork (for example, an unknown source session) still ends without raising.

strict_config(e)

@spec strict_config(t()) :: t()

Error out when config.toml contains fields this Codex version does not recognize.

supported?(config)

@spec supported?(CodexWrapper.Config.t()) :: boolean()

Return whether the installed CLI has codex exec fork.

Runs codex exec fork --help, which needs no authentication and starts no session.

validate(exec_fork)

@spec validate(t()) :: :ok | {:error, {:invalid_session_id, term()}}

Check the source session ID before it reaches the CLI.

Accepts a non-empty string with no leading or trailing whitespace, no control characters, and no leading - (which the CLI would parse as a flag). Both UUIDs and thread names pass. Anything else returns {:error, {:invalid_session_id, value}}.