Merge pull request #709 from Fail-Safe/fix/touch-agent-before-llm-call

fix: stamp last_active before LLM call to prevent mid-iteration heartbeat timeouts

crates/openfang-kernel/src/kernel.rs

@@ -5845,6 +5845,12 @@ impl KernelHandle for OpenFangKernel {
             .collect()
     }
 
+    fn touch_agent(&self, agent_id: &str) {
+        if let Ok(id) = agent_id.parse::<AgentId>() {
+            self.registry.touch(id);
+        }
+    }
+
     fn kill_agent(&self, agent_id: &str) -> Result<(), String> {
         let id: AgentId = agent_id
             .parse()

crates/openfang-kernel/src/registry.rs

@@ -256,6 +256,14 @@ impl AgentRegistry {
         Ok(())
     }
 
+    /// Touch an agent — refresh last_active without changing any other state.
+    /// Used by the agent loop to prevent heartbeat false-positives during long LLM calls.
+    pub fn touch(&self, id: AgentId) {
+        if let Some(mut entry) = self.agents.get_mut(&id) {
+            entry.last_active = chrono::Utc::now();
+        }
+    }
+
     /// Update an agent's system prompt (hot-swap, takes effect on next message).
     pub fn update_system_prompt(&self, id: AgentId, new_prompt: String) -> OpenFangResult<()> {
         let mut entry = self

crates/openfang-runtime/src/agent_loop.rs

@@ -381,6 +381,12 @@ pub async fn run_agent_loop(
             cb(LoopPhase::Thinking);
         }
 
+        // Stamp last_active before the (potentially long) LLM call so the
+        // heartbeat monitor doesn't flag us as unresponsive mid-iteration.
+        if let Some(k) = &kernel {
+            k.touch_agent(&agent_id_str);
+        }
+
         // Call LLM with retry, error classification, and circuit breaker
         let provider_name = manifest.model.provider.as_str();
         let mut response = call_with_retry(&*driver, request, Some(provider_name), None, &manifest.fallback_models).await?;

crates/openfang-runtime/src/kernel_handle.rs

@@ -238,6 +238,12 @@ pub trait KernelHandle: Send + Sync {
         Err("Channel file data send not available".to_string())
     }
 
+    /// Refresh an agent's last_active timestamp without changing any other state.
+    /// Called by the agent loop before long LLM calls to prevent heartbeat false-positives.
+    fn touch_agent(&self, agent_id: &str) {
+        let _ = agent_id;
+    }
+
     /// Spawn an agent with capability inheritance enforcement.
     /// `parent_caps` are the parent's granted capabilities. The kernel MUST verify
     /// that every capability in the child manifest is covered by `parent_caps`.