diff --git a/crates/zclaw-kernel/src/kernel/mod.rs b/crates/zclaw-kernel/src/kernel/mod.rs
index a66fe21..ed2dc61 100644
--- a/crates/zclaw-kernel/src/kernel/mod.rs
+++ b/crates/zclaw-kernel/src/kernel/mod.rs
@@ -7,6 +7,7 @@ mod skills;
 mod hands;
 mod triggers;
 mod approvals;
+mod orchestration;
 #[cfg(feature = "multi-agent")]
 mod a2a;
 
diff --git a/crates/zclaw-kernel/src/kernel/orchestration.rs b/crates/zclaw-kernel/src/kernel/orchestration.rs
new file mode 100644
index 0000000..733dae5
--- /dev/null
+++ b/crates/zclaw-kernel/src/kernel/orchestration.rs
@@ -0,0 +1,71 @@
+//! Kernel-level orchestration methods
+//!
+//! Bridges the zclaw-skills orchestration engine to the Kernel,
+//! providing high-level orchestration execution via Tauri commands.
+
+use std::collections::HashMap;
+use serde_json::Value;
+use zclaw_types::Result;
+
+use super::Kernel;
+use zclaw_skills::orchestration::{OrchestrationPlanner, SkillGraphExecutor};
+
+impl Kernel {
+    /// Execute a skill orchestration graph with the given inputs.
+    ///
+    /// This creates a `DefaultExecutor` backed by the kernel's `SkillRegistry`,
+    /// builds an execution plan, and runs the graph with true parallel execution
+    /// within each parallel group.
+    pub async fn execute_orchestration(
+        &self,
+        graph: &zclaw_skills::orchestration::SkillGraph,
+        inputs: HashMap<String, Value>,
+        context: &zclaw_skills::SkillContext,
+    ) -> Result<zclaw_skills::orchestration::OrchestrationResult> {
+        let executor = zclaw_skills::orchestration::DefaultExecutor::new(
+            self.skills.clone(),
+        );
+
+        executor.execute(graph, inputs, context).await
+    }
+
+    /// Execute an orchestration with progress tracking.
+    ///
+    /// The `progress_fn` callback is invoked at each stage of execution,
+    /// allowing the frontend to display real-time progress.
+    pub async fn execute_orchestration_with_progress<F>(
+        &self,
+        graph: &zclaw_skills::orchestration::SkillGraph,
+        inputs: HashMap<String, Value>,
+        context: &zclaw_skills::SkillContext,
+        progress_fn: F,
+    ) -> Result<zclaw_skills::orchestration::OrchestrationResult>
+    where
+        F: Fn(zclaw_skills::orchestration::OrchestrationProgress) + Send + Sync,
+    {
+        let executor = zclaw_skills::orchestration::DefaultExecutor::new(
+            self.skills.clone(),
+        );
+
+        executor.execute_with_progress(graph, inputs, context, progress_fn).await
+    }
+
+    /// Auto-compose a skill graph from a list of skill IDs.
+    ///
+    /// Analyzes input/output schemas to infer data flow edges between skills.
+    pub async fn auto_compose_skills(
+        &self,
+        skill_ids: &[zclaw_types::SkillId],
+    ) -> Result<zclaw_skills::orchestration::SkillGraph> {
+        let planner = zclaw_skills::orchestration::DefaultPlanner::new();
+        planner.auto_compose(skill_ids, &self.skills).await
+    }
+
+    /// Validate a skill graph without executing it.
+    pub async fn validate_orchestration(
+        &self,
+        graph: &zclaw_skills::orchestration::SkillGraph,
+    ) -> Vec<zclaw_skills::orchestration::ValidationError> {
+        zclaw_skills::orchestration::validate_graph(graph, &self.skills).await
+    }
+}
diff --git a/crates/zclaw-skills/src/orchestration/executor.rs b/crates/zclaw-skills/src/orchestration/executor.rs
index 36009d3..13a53db 100644
--- a/crates/zclaw-skills/src/orchestration/executor.rs
+++ b/crates/zclaw-skills/src/orchestration/executor.rs
@@ -17,6 +17,12 @@ use super::{
     planner::OrchestrationPlanner,
 };
 
+/// Wrapper to make NodeResult Send for JoinSet
+struct ParallelNodeResult {
+    node_id: String,
+    result: NodeResult,
+}
+
 /// Skill graph executor trait
 #[async_trait::async_trait]
 pub trait SkillGraphExecutor: Send + Sync {
@@ -76,7 +82,8 @@ impl DefaultExecutor {
         cancellations.get(graph_id).copied().unwrap_or(false)
     }
 
-    /// Execute a single node
+    /// Execute a single node (used by pipeline orchestration action driver)
+    #[allow(dead_code)]
     async fn execute_node(
         &self,
         node: &super::SkillNode,
@@ -201,7 +208,7 @@ impl SkillGraphExecutor for DefaultExecutor {
         let mut node_results: HashMap<String, NodeResult> = HashMap::new();
         let mut progress = OrchestrationProgress::new(&graph.id, graph.nodes.len());
 
-        // Execute parallel groups
+        // Execute parallel groups sequentially, but nodes within each group in parallel
         for group in &plan.parallel_groups {
             if self.is_cancelled(&graph.id).await {
                 return Ok(OrchestrationResult {
@@ -213,40 +220,72 @@ impl SkillGraphExecutor for DefaultExecutor {
                 });
             }
 
-            // Execute nodes in parallel within the group
+            progress.status = format!("Executing group with {} nodes", group.len());
+            progress_fn(progress.clone());
+
+            // Execute all nodes in the group concurrently using JoinSet
+            let mut join_set = tokio::task::JoinSet::new();
             for node_id in group {
                 if let Some(node) = graph.nodes.iter().find(|n| &n.id == node_id) {
-                    progress.current_node = Some(node_id.clone());
-                    progress_fn(progress.clone());
+                    let node = node.clone();
+                    let node_id = node_id.clone();
+                    let orch_ctx = orch_context.clone();
+                    let skill_ctx = context.clone();
+                    let registry = self.registry.clone();
 
-                    let result = self.execute_node(node, &orch_context, context).await
-                        .unwrap_or_else(|e| NodeResult {
-                            node_id: node_id.clone(),
-                            success: false,
-                            output: Value::Null,
-                            error: Some(e.to_string()),
-                            duration_ms: 0,
-                            retries: 0,
-                            skipped: false,
-                        });
-                    node_results.insert(node_id.clone(), result);
+                    join_set.spawn(async move {
+                        let input = orch_ctx.resolve_node_input(&node);
+                        let start = Instant::now();
+
+                        let result = registry.execute(&node.skill_id, &skill_ctx, input).await;
+                        let nr = match result {
+                            Ok(sr) if sr.success => NodeResult {
+                                node_id: node_id.clone(),
+                                success: true,
+                                output: sr.output,
+                                error: None,
+                                duration_ms: start.elapsed().as_millis() as u64,
+                                retries: 0,
+                                skipped: false,
+                            },
+                            Ok(sr) => NodeResult {
+                                node_id: node_id.clone(),
+                                success: false,
+                                output: Value::Null,
+                                error: sr.error,
+                                duration_ms: start.elapsed().as_millis() as u64,
+                                retries: 0,
+                                skipped: false,
+                            },
+                            Err(e) => NodeResult {
+                                node_id: node_id.clone(),
+                                success: false,
+                                output: Value::Null,
+                                error: Some(e.to_string()),
+                                duration_ms: start.elapsed().as_millis() as u64,
+                                retries: 0,
+                                skipped: false,
+                            },
+                        };
+                        ParallelNodeResult { node_id, result: nr }
+                    });
                 }
             }
 
-            // Update context with node outputs
-            for node_id in group {
-                if let Some(result) = node_results.get(node_id) {
-                    if result.success {
-                        orch_context.set_node_output(node_id, result.output.clone());
-                        progress.completed_nodes.push(node_id.clone());
-                    } else {
-                        progress.failed_nodes.push(node_id.clone());
-
-                        // Handle error based on strategy
-                        match graph.on_error {
-                            ErrorStrategy::Stop => {
-                                // Clone error before moving node_results
+            // Collect results as tasks complete
+            while let Some(join_result) = join_set.join_next().await {
+                match join_result {
+                    Ok(parallel_result) => {
+                        let ParallelNodeResult { node_id, result } = parallel_result;
+                        if result.success {
+                            orch_context.set_node_output(&node_id, result.output.clone());
+                            progress.completed_nodes.push(node_id.clone());
+                        } else {
+                            progress.failed_nodes.push(node_id.clone());
+                            if matches!(graph.on_error, ErrorStrategy::Stop) {
                                 let error = result.error.clone();
+                                node_results.insert(node_id, result);
+                                join_set.abort_all();
                                 return Ok(OrchestrationResult {
                                     success: false,
                                     output: Value::Null,
@@ -255,27 +294,24 @@ impl SkillGraphExecutor for DefaultExecutor {
                                     error,
                                 });
                             }
-                            ErrorStrategy::Continue => {
-                                // Continue to next group
-                            }
-                            ErrorStrategy::Retry => {
-                                // Already handled in execute_node
-                            }
                         }
+                        node_results.insert(node_id, result);
+                    }
+                    Err(e) => {
+                        tracing::warn!("[Orchestration] Task panicked: {}", e);
                     }
                 }
             }
 
             // Update progress
             progress.progress_percent = ((progress.completed_nodes.len() + progress.failed_nodes.len())
-                * 100 / graph.nodes.len()) as u8;
+                * 100 / graph.nodes.len().max(1)) as u8;
             progress.status = format!("Completed group with {} nodes", group.len());
             progress_fn(progress.clone());
         }
 
         // Build final output
         let output = orch_context.build_output(&graph.output_mapping);
-
         let success = progress.failed_nodes.is_empty();
 
         Ok(OrchestrationResult {
diff --git a/desktop/src-tauri/src/kernel_commands/mod.rs b/desktop/src-tauri/src/kernel_commands/mod.rs
index 275cf27..686c5bf 100644
--- a/desktop/src-tauri/src/kernel_commands/mod.rs
+++ b/desktop/src-tauri/src/kernel_commands/mod.rs
@@ -13,6 +13,7 @@ pub mod chat;
 pub mod hand;
 pub mod lifecycle;
 pub mod mcp;
+pub mod orchestration;
 pub mod scheduled_task;
 pub mod skill;
 pub mod trigger;
diff --git a/desktop/src-tauri/src/kernel_commands/orchestration.rs b/desktop/src-tauri/src/kernel_commands/orchestration.rs
new file mode 100644
index 0000000..dea58ad
--- /dev/null
+++ b/desktop/src-tauri/src/kernel_commands/orchestration.rs
@@ -0,0 +1,273 @@
+//! Skill orchestration Tauri commands
+//!
+//! Exposes the multi-skill orchestration engine to the frontend,
+//! enabling parallel/serial/conditional execution of skill graphs.
+
+use std::collections::HashMap;
+use serde::{Deserialize, Serialize};
+use serde_json::Value;
+use tauri::State;
+
+use super::KernelState;
+
+// ============================================================================
+// Request/Response Types
+// ============================================================================
+
+/// Orchestration execution request
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct OrchestrationRequest {
+    /// Skill IDs to orchestrate (for auto-compose)
+    pub skill_ids: Vec<String>,
+    /// Optional pre-defined graph (overrides auto-compose)
+    pub graph: Option<OrchestrationGraphInput>,
+    /// Input data for the orchestration
+    pub inputs: HashMap<String, Value>,
+    /// Agent context
+    pub agent_id: String,
+    pub session_id: String,
+}
+
+/// Graph input for pre-defined orchestration
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct OrchestrationGraphInput {
+    pub id: String,
+    pub name: String,
+    #[serde(default)]
+    pub description: String,
+    pub nodes: Vec<NodeInput>,
+    #[serde(default)]
+    pub edges: Vec<EdgeInput>,
+    #[serde(default)]
+    pub on_error: Option<String>,
+    #[serde(default)]
+    pub timeout_secs: Option<u64>,
+}
+
+/// Node input for orchestration graph
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct NodeInput {
+    pub id: String,
+    pub skill_id: String,
+    #[serde(default)]
+    pub description: String,
+    #[serde(default)]
+    pub input_mappings: HashMap<String, String>,
+    #[serde(default)]
+    pub when: Option<String>,
+    #[serde(default)]
+    pub skip_on_error: Option<bool>,
+    #[serde(default)]
+    pub timeout_secs: Option<u64>,
+}
+
+/// Edge input for orchestration graph
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct EdgeInput {
+    pub from_node: String,
+    pub to_node: String,
+    #[serde(default)]
+    pub field_mapping: HashMap<String, String>,
+    #[serde(default)]
+    pub condition: Option<String>,
+}
+
+/// Orchestration execution result for frontend
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct OrchestrationResponse {
+    pub success: bool,
+    pub output: Value,
+    pub node_count: usize,
+    pub completed_nodes: usize,
+    pub failed_nodes: usize,
+    pub duration_ms: u64,
+    pub error: Option<String>,
+    pub node_results: HashMap<String, NodeResultResponse>,
+}
+
+/// Node execution result for frontend
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct NodeResultResponse {
+    pub success: bool,
+    pub output: Value,
+    pub error: Option<String>,
+    pub duration_ms: u64,
+    pub retries: u32,
+    pub skipped: bool,
+}
+
+/// Validation response for frontend
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct ValidationResponse {
+    pub valid: bool,
+    pub errors: Vec<ValidationErrorInfo>,
+}
+
+/// Validation error info
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct ValidationErrorInfo {
+    pub code: String,
+    pub message: String,
+    pub location: Option<String>,
+}
+
+// ============================================================================
+// Type Conversions
+// ============================================================================
+
+impl TryFrom<OrchestrationGraphInput> for zclaw_skills::orchestration::SkillGraph {
+    type Error = String;
+
+    fn try_from(input: OrchestrationGraphInput) -> Result<Self, Self::Error> {
+        let nodes: Vec<zclaw_skills::orchestration::SkillNode> = input.nodes.into_iter().map(|n| {
+            zclaw_skills::orchestration::SkillNode {
+                id: n.id,
+                skill_id: zclaw_types::SkillId::new(&n.skill_id),
+                description: n.description,
+                input_mappings: n.input_mappings,
+                retry: None,
+                timeout_secs: n.timeout_secs,
+                when: n.when,
+                skip_on_error: n.skip_on_error.unwrap_or(false),
+            }
+        }).collect();
+
+        let edges: Vec<zclaw_skills::orchestration::SkillEdge> = input.edges.into_iter().map(|e| {
+            zclaw_skills::orchestration::SkillEdge {
+                from_node: e.from_node,
+                to_node: e.to_node,
+                field_mapping: e.field_mapping,
+                condition: e.condition,
+            }
+        }).collect();
+
+        let on_error = match input.on_error.as_deref() {
+            Some("continue") => zclaw_skills::orchestration::ErrorStrategy::Continue,
+            Some("retry") => zclaw_skills::orchestration::ErrorStrategy::Retry,
+            _ => zclaw_skills::orchestration::ErrorStrategy::Stop,
+        };
+
+        Ok(zclaw_skills::orchestration::SkillGraph {
+            id: input.id,
+            name: input.name,
+            description: input.description,
+            nodes,
+            edges,
+            input_schema: None,
+            output_mapping: HashMap::new(),
+            on_error,
+            timeout_secs: input.timeout_secs.unwrap_or(300),
+        })
+    }
+}
+
+impl From<zclaw_skills::orchestration::OrchestrationResult> for OrchestrationResponse {
+    fn from(result: zclaw_skills::orchestration::OrchestrationResult) -> Self {
+        let completed = result.node_results.values().filter(|r| r.success).count();
+        let failed = result.node_results.values().filter(|r| !r.success).count();
+
+        let node_results = result.node_results.into_iter().map(|(id, nr)| {
+            (id, NodeResultResponse {
+                success: nr.success,
+                output: nr.output,
+                error: nr.error,
+                duration_ms: nr.duration_ms,
+                retries: nr.retries,
+                skipped: nr.skipped,
+            })
+        }).collect();
+
+        Self {
+            success: result.success,
+            output: result.output,
+            node_count: completed + failed,
+            completed_nodes: completed,
+            failed_nodes: failed,
+            duration_ms: result.duration_ms,
+            error: result.error,
+            node_results,
+        }
+    }
+}
+
+// ============================================================================
+// Tauri Commands
+// ============================================================================
+
+/// Execute a skill orchestration
+///
+/// Either auto-composes a graph from skill_ids, or uses a pre-defined graph.
+/// Executes with true parallel execution within each dependency level.
+#[tauri::command]
+pub async fn orchestration_execute(
+    state: State<'_, KernelState>,
+    request: OrchestrationRequest,
+) -> Result<OrchestrationResponse, String> {
+    let kernel_lock = state.lock().await;
+    let kernel = kernel_lock.as_ref()
+        .ok_or_else(|| "Kernel not initialized".to_string())?;
+
+    // Build or auto-compose graph
+    let graph = if let Some(graph_input) = request.graph {
+        graph_input.try_into()?
+    } else if request.skill_ids.is_empty() {
+        return Err("Must provide either skill_ids or graph".to_string());
+    } else {
+        let skill_ids: Vec<zclaw_types::SkillId> = request.skill_ids.iter()
+            .map(|s| zclaw_types::SkillId::new(s))
+            .collect();
+
+        kernel.auto_compose_skills(&skill_ids).await
+            .map_err(|e| format!("Auto-compose failed: {}", e))?
+    };
+
+    // Build skill context
+    let context = zclaw_skills::SkillContext {
+        agent_id: request.agent_id,
+        session_id: request.session_id,
+        working_dir: None,
+        env: HashMap::new(),
+        timeout_secs: graph.timeout_secs,
+        network_allowed: true,
+        file_access_allowed: true,
+        llm: None,
+    };
+
+    // Execute orchestration
+    let result = kernel.execute_orchestration(&graph, request.inputs, &context).await
+        .map_err(|e| format!("Orchestration failed: {}", e))?;
+
+    Ok(OrchestrationResponse::from(result))
+}
+
+/// Validate an orchestration graph without executing it
+#[tauri::command]
+pub async fn orchestration_validate(
+    state: State<'_, KernelState>,
+    graph: OrchestrationGraphInput,
+) -> Result<ValidationResponse, String> {
+    let kernel_lock = state.lock().await;
+    let kernel = kernel_lock.as_ref()
+        .ok_or_else(|| "Kernel not initialized".to_string())?;
+
+    let skill_graph: zclaw_skills::orchestration::SkillGraph = graph.try_into()?;
+
+    let errors = kernel.validate_orchestration(&skill_graph).await;
+
+    let valid = errors.is_empty();
+    let error_infos = errors.into_iter().map(|e| ValidationErrorInfo {
+        code: e.code,
+        message: e.message,
+        location: e.location,
+    }).collect();
+
+    Ok(ValidationResponse { valid, errors: error_infos })
+}
diff --git a/desktop/src-tauri/src/lib.rs b/desktop/src-tauri/src/lib.rs
index 350f215..fb32ac4 100644
--- a/desktop/src-tauri/src/lib.rs
+++ b/desktop/src-tauri/src/lib.rs
@@ -146,6 +146,9 @@ pub fn run() {
             kernel_commands::skill::skill_create,
             kernel_commands::skill::skill_update,
             kernel_commands::skill::skill_delete,
+            // Orchestration commands (multi-skill graphs)
+            kernel_commands::orchestration::orchestration_execute,
+            kernel_commands::orchestration::orchestration_validate,
             // Hands commands (autonomous capabilities)
             kernel_commands::hand::hand_list,
             kernel_commands::hand::hand_execute,