fix: kill stale cargo processes before running acceptance gates

The completion handler now pgrep+kills any cargo processes targeting
the worktree's Cargo.toml before running gates. This prevents the
run_tests MCP child from holding the build lock and blocking gates.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

server/src/agents/pool/pipeline/completion.rs

@@ -199,6 +199,25 @@ pub(in crate::agents::pool) async fn run_server_owned_completion(
             .and_then(|a| a.worktree_info.as_ref().map(|wt| wt.path.clone()))
     };
 
+    // Kill any in-flight cargo test processes for this worktree so they don't
+    // hold the build lock while gates try to run.
+    if let Some(wt_path) = worktree_path.as_ref() {
+        if let Ok(output) = std::process::Command::new("pgrep")
+            .args(["-f", &format!("--manifest-path {}/Cargo.toml", wt_path.display())])
+            .output()
+        {
+            let pids = String::from_utf8_lossy(&output.stdout);
+            for pid_str in pids.lines() {
+                if let Ok(pid) = pid_str.trim().parse::<i32>() {
+                    crate::slog!(
+                        "[agents] Killing stale cargo process (pid {pid}) for '{story_id}' before running gates"
+                    );
+                    unsafe { libc::kill(pid, libc::SIGKILL); }
+                }
+            }
+        }
+    }
+
     // Run acceptance gates.
     let (gates_passed, gate_output) = if let Some(wt_path) = worktree_path {
         let path = wt_path;