Struct ClaudeCodeExecutor 

pub struct ClaudeCodeExecutor { /* private fields */ }

Drives claude --output-format stream-json --input-format stream-json ... as the engine behind one sub-agent run.

Implementations

impl ClaudeCodeExecutor

pub fn new( binary: Option<String>, model: Option<String>, permission_mode: Option<String>, workspace: Option<String>, state_dir: Option<PathBuf>, inherit_user_config: bool, forward_env: Vec<String>, ) -> Self

Trait Implementations

impl ChildExecutor for ClaudeCodeExecutor

fn run<'life0, 'async_trait>( &'life0 self, spec: RunSpec, events: EventSink, steer: SteerInbox, cancel: CancellationToken, ) -> Pin<Box<dyn Future<Output = ChildOutcome> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Activation logic (issue #444), driven by spec.messages — empty means first activation, non-empty means a reactivation carrying prior context (RunSpec.messages doc, proto.rs:28):

1. messages empty → fresh session; delete any stale state file (a rerun must never accidentally resume).
2. messages non-empty AND the state file has an id recorded against the SAME workspace → spawn with --resume <id>, sending just the live assignment (the CLI already has the transcript).
3. messages non-empty but no usable id (first run on this machine, storage GC’d, workspace changed) → fallback: render the shipped history into a bounded preamble prepended to the assignment.
4. If a --resume spawn exits without ever producing a result frame (bad/GC’d session id — the CLI errors out fast), retry ONCE without --resume, using the same fallback rehydration as step 3. No retry loop beyond this single attempt.