ci: unblock task branch checks

.github/workflows/ci-firecracker.yml

@@ -42,10 +42,30 @@ jobs:
       - name: Create VM
         id: vm
         run: |
-          RESPONSE=$(curl -sf -X POST $FCCTL_URL/api/vms \
+          RESPONSE_FILE=$(mktemp)
+          trap 'rm -f "${RESPONSE_FILE}"' EXIT
+          if HTTP_STATUS=$(curl -sS -o "$RESPONSE_FILE" -w "%{http_code}" -X POST "$FCCTL_URL/api/vms" \
             -H 'Content-Type: application/json' \
-            -d "{\"vm_type\": \"$VM_TYPE\"}")
+            -d "{\"vm_type\": \"$VM_TYPE\"}"); then
+            :
+          else
+            CURL_EXIT=$?
+            RESPONSE=$(cat "$RESPONSE_FILE")
+            echo "fcctl create transport failed: curl exit $CURL_EXIT"
+            if [ -n "$RESPONSE" ]; then
+              echo "$RESPONSE"
+            fi
+            exit 1
+          fi
+          RESPONSE=$(cat "$RESPONSE_FILE")
+          rm -f "$RESPONSE_FILE"
+          trap - EXIT
+          echo "fcctl create status: $HTTP_STATUS"
           echo "$RESPONSE"
+          if [ "$HTTP_STATUS" -lt 200 ] || [ "$HTTP_STATUS" -ge 300 ]; then
+            echo "ERROR: fcctl-web VM create failed with HTTP $HTTP_STATUS"
+            exit 1
+          fi
           VM_ID=$(echo "$RESPONSE" | python3 -c "import sys,json; print(json.load(sys.stdin)['id'])")
           if [ -z "$VM_ID" ] || [ "$VM_ID" = "null" ]; then
             echo "ERROR: failed to parse vm id from response"
@@ -144,8 +164,8 @@ jobs:
       # bigbox; see .cargo/config.toml for the Darwin `dynamic_lookup`
       # linker workaround on local Mac dev.
       #
-      # All cargo invocations are dispatched via `rch exec --` so they
-      # share rchd's queue + slot accounting with ADF agents (see
+      # Build-oriented cargo invocations are dispatched via `rch exec --` so
+      # they share rchd's queue + slot accounting with ADF agents (see
       # .docs/adr-rch-build-queue-not-firecracker-ci.md). Fail-open: if
       # rchd is down or no slot is available, rch falls through to local
       # cargo with no behaviour change.
@@ -161,15 +181,34 @@ jobs:
       - name: Install cargo-nextest
         run: |
           if ! command -v cargo-nextest >/dev/null 2>&1; then
-            cargo install cargo-nextest --locked
+            curl -LsSf https://get.nexte.st/latest/linux | tar zxf - -C "${CARGO_HOME:-$HOME/.cargo}/bin"
           fi
           cargo nextest --version
 
       - name: cargo nextest run --workspace
-        # Only test_chat_command is skipped: it requires LLM API credentials
-        # not present in CI. All other failures must be fixed at the source,
-        # not skipped. nextest uses filter expressions instead of --skip.
-        run: /home/alex/.local/bin/rch exec -- cargo nextest run --workspace --profile ci -E 'not test(test_chat_command)'
+        # test_chat_command is skipped: it requires LLM API credentials not
+        # present in CI. The MCP autocomplete E2E binary is excluded here only
+        # because it spawns nested `cargo run`; the following step covers it
+        # serially to avoid build-dir lock contention. All other failures must
+        # be fixed at the source, not skipped. nextest uses filter expressions
+        # instead of --skip.
+        # Do not wrap these invocations in `rch exec`: rch's non-compilation
+        # command path loses shell quoting around filter expressions and makes
+        # `/bin/sh` parse parentheses in predicates like `test(...)`.
+        # `RUSTC_WRAPPER=sccache` above still applies to rustc invocations.
+        run: |
+          cargo nextest run --workspace --profile ci \
+            -E 'not test(test_chat_command) - binary_id(terraphim_mcp_server::mcp_autocomplete_e2e_test)'
+
+      - name: cargo nextest run mcp_autocomplete_e2e serially
+        # These tests spawn `cargo run --bin terraphim_mcp_server`. Under
+        # nextest each test is its own process, so `#[serial]` cannot prevent
+        # concurrent nested Cargo invocations. Run this binary separately with
+        # one test thread to avoid build-dir lock contention and CI timeouts.
+        run: |
+          cargo nextest run -p terraphim_mcp_server --test mcp_autocomplete_e2e_test \
+            --profile ci --test-threads 1 --no-capture \
+            -E 'not test(test_chat_command)'
 
       - name: sccache stats
         if: always()

.github/workflows/performance-benchmarking.yml

@@ -38,7 +38,7 @@ jobs:
   performance-benchmarks:
     name: Performance Benchmarks
     runs-on: ubuntu-latest
-    timeout-minutes: 30
+    timeout-minutes: 45
 
     steps:
     - name: Checkout code

crates/terraphim_orchestrator/src/lib_tests.rs

@@ -193,11 +193,14 @@ async fn test_direct_dispatch_config_starts_socket_listener() {
 
 #[tokio::test]
 async fn test_handle_direct_dispatch_spawns_agent_without_mentions() {
+    let temp = TempDir::new().unwrap();
     let mut config = test_config();
+    config.working_dir = temp.path().to_path_buf();
+    config.compound_review.worktree_root = temp.path().join(".worktrees");
     config.agents = vec![AgentDefinition {
         name: "echo-agent".to_string(),
         layer: AgentLayer::Core,
-        cli_tool: "echo".to_string(),
+        cli_tool: "/bin/echo".to_string(),
         task: "echo hello".to_string(),
         schedule: None,
         model: None,

crates/terraphim_rlm/src/main.rs

@@ -48,7 +48,10 @@ async fn run(cli: Cli) -> Result<CliResponse, Box<dyn std::error::Error>> {
     #[cfg(feature = "llm")]
     {
         if let Err(e) = rlm.auto_configure_llm().await {
-            log::warn!("LLM auto-configuration failed: {}. rlm_query will be unavailable.", e);
+            log::warn!(
+                "LLM auto-configuration failed: {}. rlm_query will be unavailable.",
+                e
+            );
         }
     }
 

crates/terraphim_rlm/src/query_loop.rs

@@ -690,7 +690,12 @@ fn truncate(s: &str, max_len: usize) -> String {
     if s.len() <= max_len {
         s.to_string()
     } else {
-        let boundary = s.floor_char_boundary(max_len);
+        let boundary = s
+            .char_indices()
+            .map(|(idx, _)| idx)
+            .take_while(|idx| *idx <= max_len)
+            .last()
+            .unwrap_or(0);
         format!("{}...", &s[..boundary])
     }
 }

crates/terraphim_rlm/src/rlm.rs

@@ -910,8 +910,7 @@ impl TerraphimRlm {
             "llm_provider".to_string(),
             serde_json::Value::String("ollama".to_string()),
         );
-        let ollama_model = std::env::var("RLM_MODEL")
-            .unwrap_or_else(|_| "gemma3:270m".to_string());
+        let ollama_model = std::env::var("RLM_MODEL").unwrap_or_else(|_| "gemma3:270m".to_string());
         role.extra.insert(
             "llm_model".to_string(),
             serde_json::Value::String(ollama_model.clone()),
@@ -929,7 +928,9 @@ impl TerraphimRlm {
                 ])
                 .output()
                 .ok()?;
-            String::from_utf8(output.stdout).ok().map(|s| s.trim().to_string())
+            String::from_utf8(output.stdout)
+                .ok()
+                .map(|s| s.trim().to_string())
         });
 
         if let Some(ref key) = or_api_key {
@@ -938,7 +939,9 @@ impl TerraphimRlm {
             role.llm_api_key = Some(key.clone());
             role.llm_model = Some(or_model.clone());
             // Cache for child processes and build_llm_from_role
-            unsafe { std::env::set_var("OPENROUTER_API_KEY", key); }
+            unsafe {
+                std::env::set_var("OPENROUTER_API_KEY", key);
+            }
             log::info!("RLM auto-configure: openrouter model={}", or_model);
         }
 
@@ -958,11 +961,10 @@ impl TerraphimRlm {
             ..Default::default()
         });
 
-        let client = terraphim_service::llm::build_llm_from_role(&role)
-            .ok_or_else(|| {
-                log::warn!("RLM auto-configure: no LLM provider available");
-                RlmError::LlmNotConfigured
-            })?;
+        let client = terraphim_service::llm::build_llm_from_role(&role).ok_or_else(|| {
+            log::warn!("RLM auto-configure: no LLM provider available");
+            RlmError::LlmNotConfigured
+        })?;
 
         log::info!(
             "RLM LLM bridge configured: providers={} strategy={:?}",

crates/terraphim_rlm/tests/backend_demo.rs

@@ -1,8 +1,8 @@
 //! Demonstration: RLM running locally (LocalExecutor) and via Docker (DockerExecutor).
 //! Run: cargo test -p terraphim_rlm --test backend_demo -- --nocapture
 
-use terraphim_rlm::config::{BackendType, RlmConfig};
 use terraphim_rlm::TerraphimRlm;
+use terraphim_rlm::config::{BackendType, RlmConfig};
 
 #[tokio::test]
 async fn demo_local_executor() {
@@ -19,11 +19,22 @@ async fn demo_local_executor() {
 
     // Python
     let r = rlm.execute_code(&session.id, "print(2+2)").await.unwrap();
-    println!("  [Python] 2+2 = {} (exit {})", r.stdout.trim(), r.exit_code);
+    println!(
+        "  [Python] 2+2 = {} (exit {})",
+        r.stdout.trim(),
+        r.exit_code
+    );
 
     // Bash
-    let r = rlm.execute_command(&session.id, "echo hello-from-local").await.unwrap();
-    println!("  [Bash]   echo = {} (exit {})", r.stdout.trim(), r.exit_code);
+    let r = rlm
+        .execute_command(&session.id, "echo hello-from-local")
+        .await
+        .unwrap();
+    println!(
+        "  [Bash]   echo = {} (exit {})",
+        r.stdout.trim(),
+        r.exit_code
+    );
 
     // Show backend type
     let status = rlm.get_session_status(&session.id, false).await.unwrap();
@@ -55,11 +66,22 @@ async fn demo_docker_executor() {
 
     // Python
     let r = rlm.execute_code(&session.id, "print(2+2)").await.unwrap();
-    println!("  [Python] 2+2 = {} (exit {})", r.stdout.trim(), r.exit_code);
+    println!(
+        "  [Python] 2+2 = {} (exit {})",
+        r.stdout.trim(),
+        r.exit_code
+    );
 
     // Bash
-    let r = rlm.execute_command(&session.id, "echo hello-from-docker").await.unwrap();
-    println!("  [Bash]   echo = {} (exit {})", r.stdout.trim(), r.exit_code);
+    let r = rlm
+        .execute_command(&session.id, "echo hello-from-docker")
+        .await
+        .unwrap();
+    println!(
+        "  [Bash]   echo = {} (exit {})",
+        r.stdout.trim(),
+        r.exit_code
+    );
 
     // Show backend type
     let status = rlm.get_session_status(&session.id, false).await.unwrap();
@@ -73,11 +95,17 @@ async fn demo_docker_executor() {
     println!("  [Container hostname] {}", r.stdout.trim());
 
     // Show Python version inside container
-    let r = rlm.execute_code(&session.id, "import sys; print(sys.version)").await.unwrap();
+    let r = rlm
+        .execute_code(&session.id, "import sys; print(sys.version)")
+        .await
+        .unwrap();
     println!("  [Python version] {}", r.stdout.trim());
 
     // Show container filesystem
-    let r = rlm.execute_command(&session.id, "ls / | head -5").await.unwrap();
+    let r = rlm
+        .execute_command(&session.id, "ls / | head -5")
+        .await
+        .unwrap();
     println!("  [Container root]\n{}", r.stdout);
 
     rlm.destroy_session(&session.id).await.unwrap();

crates/terraphim_rlm/tests/skills_demo.rs

@@ -6,21 +6,17 @@ use terraphim_rlm::{RlmConfig, TerraphimRlm};
 #[tokio::test]
 async fn demo_all_skills() {
     let config = RlmConfig::minimal();
-    let rlm = TerraphimRlm::with_executor(
-        config,
-        terraphim_rlm::LocalExecutor::new(),
-    )
-    .unwrap();
+    let rlm = TerraphimRlm::with_executor(config, terraphim_rlm::LocalExecutor::new()).unwrap();
 
     // 1. Session create
     let session = rlm.create_session().await.unwrap();
-    println!("[session create] id={} state={:?}", session.id, session.state);
+    println!(
+        "[session create] id={} state={:?}",
+        session.id, session.state
+    );
 
     // 2. Code execution
-    let result = rlm
-        .execute_code(&session.id, "print(2+2)")
-        .await
-        .unwrap();
+    let result = rlm.execute_code(&session.id, "print(2+2)").await.unwrap();
     println!(
         "[code: 2+2] exit={} stdout={:?}",
         result.exit_code, result.stdout
@@ -64,10 +60,7 @@ async fn demo_all_skills() {
     assert_eq!(val, None);
 
     // 8. Status
-    let status = rlm
-        .get_session_status(&session.id, false)
-        .await
-        .unwrap();
+    let status = rlm.get_session_status(&session.id, false).await.unwrap();
     println!(
         "[status] backend={:?} snapshots={}",
         status.backend_type, status.snapshot_count

terraphim_server/tests/api_context_tests.rs

@@ -14,7 +14,6 @@ use terraphim_server::{
     GetConversationResponse, ListConversationsResponse, Status, axum_server,
 };
 use terraphim_service::http_client;
-use terraphim_settings::DeviceSettings;
 use terraphim_types::{ContextType, Document, DocumentType, RelevanceFunction};
 
 /// Sample configuration for testing context management
@@ -50,15 +49,8 @@ fn create_test_config() -> Config {
 
 /// Start a test server with context management API
 async fn start_test_server() -> SocketAddr {
-    let server_settings =
-        DeviceSettings::load_from_env_and_file(None).expect("Failed to load settings");
-    let server_hostname = server_settings
-        .server_hostname
-        .parse::<SocketAddr>()
-        .unwrap_or_else(|_| {
-            let port = portpicker::pick_unused_port().expect("Failed to find unused port");
-            SocketAddr::from(([127, 0, 0, 1], port))
-        });
+    let port = portpicker::pick_unused_port().expect("Failed to find unused port");
+    let server_hostname = SocketAddr::from(([127, 0, 0, 1], port));
 
     let mut config = create_test_config();
     let config_state = terraphim_config::ConfigState::new(&mut config)