Function ingest_cursor_transcript_event_capped

src/sessions/cursor.rs:395–428

Like [`ingest_cursor_transcript_event`], but bounds how many newly-appended bytes a single call will read. Cursor hooks pass byte caps to stay within hook budgets; capped reads still discover subagent transcript files, with each file independently subject to the same cap.

(
    event_json: &str,
    db: &GlobalDb,
    max_new_bytes: Option<u64>,
)

/// budgets; capped reads still discover subagent transcript files, with each file
/// independently subject to the same cap.
pub async fn ingest_cursor_transcript_event_capped(
    event_json: &str,
    db: &GlobalDb,
    max_new_bytes: Option<u64>,
) -> CursorTranscriptIngestStats {
    let Ok(event) = serde_json::from_str::<Value>(event_json) else {
        return CursorTranscriptIngestStats::default();
    };
    let Some(transcript_path) = event
        .get("transcript_path")
        .and_then(Value::as_str)
        .filter(|path| !path.is_empty())
        .map(PathBuf::from)
    else {
        return CursorTranscriptIngestStats::default();
    };

    // Cursor derives its project from the event, so the driver's project_root
    // argument is unused by `CursorEventSource`; the transcript path's parent is
    // a cheap, side-effect-free placeholder.
    let project_root = transcript_path
        .parent()
        .map_or_else(|| transcript_path.clone(), Path::to_path_buf);
    let source = CursorEventSource {
        event,
        transcript_path,
        include_subagents: true,
    };
    let stats = ingest_source(db, &source, &project_root, max_new_bytes).await;
    CursorTranscriptIngestStats {
        sessions_upserted: stats.sessions_upserted,
        messages_upserted: stats.messages_upserted,
    }
}

/// `agent-transcripts/<session>/subagents/<child>.jsonl` is the deepest layout
/// Cursor writes; a little headroom tolerates future nesting.