sync: 0.2.2 Task subagent kind, skill inject budget, claude 2.1.236

.release-please-manifest.json

@@ -1 +1 @@
-{".":"0.2.1"}
+{".":"0.2.2"}

CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## [0.2.2](https://github.com/suoak/AionCore/compare/v0.2.1...v0.2.2) (2026-08-29)
+
+### Features
+
+* **conversation:** mid-turn interjection — deliver messages into a running claude/codex turn
+* **session:** distinguish Task subagents from background tasks (`AgentContainer`)
+* **skills:** slim auto-inject skill descriptions to the 200-char injection budget
+
+### Bug Fixes
+
+* **ai-agent:** wait for terminal output before asserting after process exit
+
+### Chores
+
+* **cli:** verify claude against 2.1.236
+* **acp:** bump codebuddy 2.141.0 and dimcode 0.3.22 registry pins
+
 ## [0.2.1](https://github.com/suoak/AionCore/compare/v0.2.0...v0.2.1) (2026-08-29)
 
 ### Bug Fixes

Cargo.toml

@@ -31,7 +31,7 @@ members = [
 ]
 
 [workspace.package]
-version = "0.2.1"
+version = "0.2.2"
 edition = "2024"
 license = "MIT"
 

crates/aionui-ai-agent/src/agent_task.rs

@@ -74,6 +74,16 @@ pub trait IAgentTask: Send + Sync {
         PromptMediaCaps::default()
     }
 
+    /// Whether a message sent right now reaches the agent without waiting for
+    /// the current turn to end (task-1 brief: mid-turn interjection). Mirrors
+    /// `aionui_session::Capabilities::supports_midturn_delivery` — defaults
+    /// false (ACP-like) so only backends that genuinely deliver mid-turn
+    /// (the clean-slate `Session` variant reading claude/codex capabilities)
+    /// opt in.
+    fn supports_midturn_delivery(&self) -> bool {
+        false
+    }
+
     /// Send a user message to the agent. Returns once the agent has
     /// accepted the turn; actual streaming proceeds on the broadcast
     /// channel returned by [`Self::subscribe`].
@@ -132,6 +142,11 @@ pub trait IMockAgent: IAgentTask {
     fn get_session_key(&self) -> Option<String> {
         None
     }
+    /// B5 mid-turn delivery test seam. Default: forward to `send_message` so
+    /// simple mocks behave; mid-turn tests override to record the routed call.
+    async fn deliver_midturn(&self, data: SendMessageData) -> Result<(), AgentSendError> {
+        self.send_message(data).await
+    }
     async fn mode(&self) -> Result<aionui_api_types::AgentModeResponse, AgentError> {
         Ok(aionui_api_types::AgentModeResponse {
             mode: "default".into(),
@@ -273,6 +288,12 @@ impl AgentInstance {
         self.as_task().subscribe()
     }
 
+    /// Whether a message sent right now reaches the agent without waiting
+    /// for the current turn to end. See `IAgentTask::supports_midturn_delivery`.
+    pub fn supports_midturn_delivery(&self) -> bool {
+        self.as_task().supports_midturn_delivery()
+    }
+
     /// Send a user message to the agent.
     pub async fn send_message(&self, data: SendMessageData) -> Result<(), AgentSendError> {
         self.as_task().send_message(data).await
@@ -308,6 +329,22 @@ impl AgentInstance {
         }
     }
 
+    /// B5 mid-turn delivery: hand a message to the RUNNING turn instead of
+    /// opening a new one. Only meaningful when
+    /// [`Self::supports_midturn_delivery`] is true — the conversation layer
+    /// gates on that bit before routing here. Variants without the path reject
+    /// (the caller must not have routed them here).
+    pub async fn deliver_midturn(&self, data: SendMessageData) -> Result<(), AgentSendError> {
+        match self {
+            Self::Acp(_) | Self::Aionrs(_) => Err(AgentSendError::from_agent_error(AgentError::BadRequest(
+                "mid-turn delivery is not supported by this agent".into(),
+            ))),
+            Self::Session(m) => m.deliver_midturn(data).await,
+            #[cfg(any(test, feature = "test-support"))]
+            Self::Mock(m) => m.deliver_midturn(data).await,
+        }
+    }
+
     /// Cancel the current streaming response without killing the agent.
     pub async fn cancel(&self) -> Result<(), AgentError> {
         self.as_task().cancel().await

crates/aionui-ai-agent/src/capability/backend_output_sink.rs

@@ -76,6 +76,7 @@ impl OutputSink for BackendOutputSink {
             input: Some(parsed_input),
             output: None,
             description: None,
+            parent_call_id: None,
         }));
     }
 
@@ -115,6 +116,7 @@ impl OutputSink for BackendOutputSink {
                 Some(content.to_owned())
             },
             description: None,
+            parent_call_id: None,
         }));
     }
 

crates/aionui-ai-agent/src/capability/backend_protocol_sink.rs

@@ -158,6 +158,7 @@ impl ProtocolEmitter for BackendProtocolSink {
                     input: None,
                     output: None,
                     description: Some(Self::execution_description(execution_id, "execute")),
+                    parent_call_id: None,
                 }));
             }
 
@@ -189,6 +190,7 @@ impl ProtocolEmitter for BackendProtocolSink {
                         error_code,
                         truncation,
                     )),
+                    parent_call_id: None,
                 }));
             }
 
@@ -219,6 +221,7 @@ impl ProtocolEmitter for BackendProtocolSink {
                         })
                         .to_string(),
                     ),
+                    parent_call_id: None,
                 }));
             }
 

crates/aionui-ai-agent/src/claude_flags.rs

@@ -126,7 +126,7 @@ mod tests {
         assert!(!supported("2.1.0"), "2.1.0 hard-errors on --thinking-display");
         assert!(!supported("2.1.190"), "below the verified floor stays off");
         assert!(supported("2.1.191"), "lowest live-probed OK version");
-        assert!(supported("2.1.235"), "the release this integration is verified against");
+        assert!(supported("2.1.236"), "the release this integration is verified against");
         assert!(supported("2.1.220"));
         assert!(supported("3.0.0"), "a future major must not regress the gate");
     }