From afbd04b5d62a890984764cc67efcb2c4ff67b4fe Mon Sep 17 00:00:00 2001
From: Lifei Zhou <lifei@squareup.com>
Date: Tue, 27 Jan 2026 22:48:40 +1100
Subject: [PATCH 1/7] debug

---
 crates/goose/src/agents/subagent_handler.rs   |  18 +
 .../goose/src/providers/formats/anthropic.rs  |  16 +
 scripts/test_compaction.sh                    | 413 ++++++++++--------
 3 files changed, 259 insertions(+), 188 deletions(-)

diff --git a/crates/goose/src/agents/subagent_handler.rs b/crates/goose/src/agents/subagent_handler.rs
index 54e0d2dab827..c2683421894d 100644
--- a/crates/goose/src/agents/subagent_handler.rs
+++ b/crates/goose/src/agents/subagent_handler.rs
@@ -186,7 +186,22 @@ fn get_agent_messages(
             .reply(user_message, session_config, cancellation_token)
             .await
             .map_err(|e| anyhow!("Failed to get reply from agent: {}", e))?;
+
+        // Stream diagnostics for debugging hangs
+        let mut event_count = 0u64;
+        let stream_start = std::time::Instant::now();
+        let mut last_event_time = stream_start;
+        tracing::info!(target: "stream_diag", "[STREAM] subagent: starting to consume agent events");
+
         while let Some(message_result) = stream.next().await {
+            event_count += 1;
+            let now = std::time::Instant::now();
+            let since_last = now.duration_since(last_event_time);
+            if since_last.as_secs() >= 5 {
+                tracing::warn!(target: "stream_diag", "[STREAM] subagent: event #{} after {:.1}s delay", event_count, since_last.as_secs_f64());
+            }
+            last_event_time = now;
+
             match message_result {
                 Ok(AgentEvent::Message(msg)) => conversation.push(msg),
                 Ok(AgentEvent::McpNotification(_)) | Ok(AgentEvent::ModelChange { .. }) => {}
@@ -200,6 +215,9 @@ fn get_agent_messages(
             }
         }
 
+        let total_duration = stream_start.elapsed();
+        tracing::info!(target: "stream_diag", "[STREAM] subagent: stream ended after {} events in {:.1}s", event_count, total_duration.as_secs_f64());
+
         let final_output = if has_response_schema {
             agent
                 .final_output_tool
diff --git a/crates/goose/src/providers/formats/anthropic.rs b/crates/goose/src/providers/formats/anthropic.rs
index 7e3dd161515a..61d6622563ff 100644
--- a/crates/goose/src/providers/formats/anthropic.rs
+++ b/crates/goose/src/providers/formats/anthropic.rs
@@ -494,7 +494,20 @@ where
         let mut final_usage: Option<crate::providers::base::ProviderUsage> = None;
         let mut message_id: Option<String> = None;
 
+        // Stream diagnostics for debugging hangs
+        let mut chunk_count = 0u64;
+        let stream_start = std::time::Instant::now();
+        let mut last_chunk_time = stream_start;
+        tracing::info!(target: "stream_diag", "[STREAM] anthropic: starting stream consumption");
+
         while let Some(line_result) = stream.next().await {
+            chunk_count += 1;
+            let now = std::time::Instant::now();
+            let since_last = now.duration_since(last_chunk_time);
+            if since_last.as_secs() >= 5 {
+                tracing::warn!(target: "stream_diag", "[STREAM] anthropic: chunk #{} after {:.1}s delay", chunk_count, since_last.as_secs_f64());
+            }
+            last_chunk_time = now;
             let line = line_result?;
 
             // Skip empty lines and non-data lines
@@ -696,6 +709,9 @@ where
             }
         }
 
+        let total_duration = stream_start.elapsed();
+        tracing::info!(target: "stream_diag", "[STREAM] anthropic: stream ended after {} chunks in {:.1}s", chunk_count, total_duration.as_secs_f64());
+
         // Yield final usage information if available
         if let Some(usage) = final_usage {
             yield (None, Some(usage));
diff --git a/scripts/test_compaction.sh b/scripts/test_compaction.sh
index baa3b7ca9660..ef95556bcb3f 100755
--- a/scripts/test_compaction.sh
+++ b/scripts/test_compaction.sh
@@ -19,6 +19,43 @@ fi
 SCRIPT_DIR=$(pwd)
 GOOSE_BIN="$SCRIPT_DIR/target/release/goose"
 
+# Timeout for goose commands (default 5 minutes)
+GOOSE_TIMEOUT=${GOOSE_TIMEOUT:-300}
+
+# Log directory for diagnostics
+LOG_DIR="$HOME/.config/goose/logs/cli"
+
+# Helper to dump stream diagnostics from logs on timeout
+dump_stream_diagnostics() {
+  echo ""
+  echo "=== STREAM DIAGNOSTICS (last 50 entries) ==="
+  if [ -d "$LOG_DIR" ]; then
+    # Find logs from the last 10 minutes and grep for stream_diag
+    find "$LOG_DIR" -name "*.log" -mmin -10 -exec grep -h "stream_diag" {} \; 2>/dev/null | tail -50
+  else
+    echo "No log directory found at $LOG_DIR"
+  fi
+  echo "=== END STREAM DIAGNOSTICS ==="
+  echo ""
+}
+
+# Helper to run goose with timeout and stream diagnostics enabled
+run_goose() {
+  local exit_code=0
+  # Enable stream_diag target for our diagnostic logs
+  RUST_LOG="${RUST_LOG:-stream_diag=info}" timeout "$GOOSE_TIMEOUT" "$GOOSE_BIN" "$@" 2>&1 || exit_code=$?
+
+  if [ $exit_code -eq 124 ] || [ $exit_code -eq 137 ]; then
+    echo ""
+    echo "⚠️  TIMEOUT: goose command exceeded ${GOOSE_TIMEOUT}s limit"
+  fi
+
+  # Always dump stream diagnostics to help debug issues
+  dump_stream_diagnostics
+
+  return $exit_code
+}
+
 # Validation function to check compaction structure in session JSON
 validate_compaction() {
   local session_id=$1
@@ -102,7 +139,7 @@ echo ""
 OUTPUT=$(mktemp)
 
 echo "Step 1: Creating session with initial messages..."
-(cd "$TESTDIR" && "$GOOSE_BIN" run --text "list files and read hello.txt" 2>&1) | tee "$OUTPUT"
+(cd "$TESTDIR" && run_goose run --text "list files and read hello.txt") | tee "$OUTPUT"
 
 if ! command -v jq &> /dev/null; then
   echo "✗ FAILED: jq is required for this test"
@@ -121,7 +158,7 @@ else
     echo "Step 2: Sending manual compaction trigger..."
 
     # Send the manual compact trigger prompt
-    (cd "$TESTDIR" && "$GOOSE_BIN" run --resume --session-id "$SESSION_ID" --text "Please compact this conversation" 2>&1) | tee -a "$OUTPUT"
+    (cd "$TESTDIR" && run_goose run --resume --session-id "$SESSION_ID" --text "Please compact this conversation") | tee -a "$OUTPUT"
 
     echo ""
     echo "Checking for compaction evidence..."
@@ -147,192 +184,192 @@ fi
 echo ""
 echo ""
 
-# ==================================================
-# TEST 2: Auto Compaction
-# ==================================================
-echo "---------------------------------------------------"
-echo "TEST 2: Auto Compaction via threshold (0.005)"
-echo "---------------------------------------------------"
-
-TESTDIR=$(mktemp -d)
-echo "test content" > "$TESTDIR/test.txt"
-echo "Test directory: $TESTDIR"
-echo ""
-
-# Set auto-compact threshold very low (.5%) to trigger it quickly
-export GOOSE_AUTO_COMPACT_THRESHOLD=0.005
-
-OUTPUT=$(mktemp)
-
-LONG_RESPONSE_PROMPT="Count from 1 to 200, one number per line."
-
-echo "Step 1: Creating session with first message (generating tokens for threshold)..."
-(cd "$TESTDIR" && "$GOOSE_BIN" run --text "$LONG_RESPONSE_PROMPT" 2>&1) | tee "$OUTPUT"
-
-if ! command -v jq &> /dev/null; then
-  echo "✗ FAILED: jq is required for this test"
-  RESULTS+=("✗ Auto Compaction (jq required)")
-else
-  SESSION_ID=$("$GOOSE_BIN" session list --format json 2>/dev/null | jq -r '.[0].id' 2>/dev/null)
-
-  if [ -z "$SESSION_ID" ] || [ "$SESSION_ID" = "null" ]; then
-    echo "✗ FAILED: Could not create session"
-    RESULTS+=("✗ Auto Compaction (no session)")
-  else
-    echo ""
-    echo "Session created: $SESSION_ID"
-    echo "Step 2: Sending second message (should trigger auto-compact)..."
-
-    # Send second message - auto-compaction should trigger before processing this
-    (cd "$TESTDIR" && "$GOOSE_BIN" run --resume --session-id "$SESSION_ID" --text "hi again" 2>&1) | tee -a "$OUTPUT"
-
-    echo ""
-    echo "Checking for auto-compaction evidence..."
-
-    if grep -qi "auto.*compact\|exceeded.*auto.*compact.*threshold" "$OUTPUT"; then
-      echo "✓ SUCCESS: Auto compaction was triggered"
-
-      if validate_compaction "$SESSION_ID" "auto compaction"; then
-        RESULTS+=("✓ Auto Compaction")
-      else
-        RESULTS+=("✗ Auto Compaction (structure validation failed)")
-      fi
-    else
-      echo "✗ FAILED: Auto compaction was not triggered"
-      echo "   Expected to see auto-compact messages with threshold of 0.005"
-      RESULTS+=("✗ Auto Compaction")
-    fi
-  fi
-fi
-
-# Unset the env variable
-unset GOOSE_AUTO_COMPACT_THRESHOLD
-
-rm -f "$OUTPUT"
-rm -rf "$TESTDIR"
-
-echo ""
-echo ""
-
-# ==================================================
-# TEST 3: Out-of-Context Error Compaction
-# ==================================================
-echo "---------------------------------------------------"
-echo "TEST 3: Compaction via out-of-context error (proxy)"
-echo "---------------------------------------------------"
-
-TESTDIR=$(mktemp -d)
-echo "test content" > "$TESTDIR/test.txt"
-echo "Test directory: $TESTDIR"
-echo ""
-
-# Use a random port to avoid conflicts
-PROXY_PORT=$((9000 + RANDOM % 1000))
-PROXY_DIR="$SCRIPT_DIR/scripts/provider-error-proxy"
-
-OUTPUT=$(mktemp)
-PROXY_LOG=$(mktemp)
-PROXY_SETUP_LOG=$(mktemp)
-
-# Pre-install proxy dependencies (so first run doesn't take forever)
-echo "Installing proxy dependencies..."
-export UV_INDEX_URL="https://pypi.org/simple"
-if ! (cd "$PROXY_DIR" && uv sync 2>&1 | tee "$PROXY_SETUP_LOG"); then
-  echo "✗ FAILED: Could not install proxy dependencies"
-  echo "Setup log:"
-  cat "$PROXY_SETUP_LOG"
-  RESULTS+=("✗ Out-of-Context Error (dependency install failed)")
-else
-  echo "✓ Dependencies installed"
-
-  # Start the error proxy in context-length error mode (3 errors)
-  echo "Starting error proxy on port $PROXY_PORT with context-length error mode..."
-  (cd "$PROXY_DIR" && UV_INDEX_URL="https://pypi.org/simple" uv run proxy.py --port "$PROXY_PORT" --mode "c 3" --no-stdin > "$PROXY_LOG" 2>&1) &
-  PROXY_PID=$!
-
-  # Wait for proxy to be ready (check if port is listening)
-  echo "Waiting for proxy to be ready..."
-  PROXY_READY=false
-  for i in {1..60}; do
-    if kill -0 $PROXY_PID 2>/dev/null; then
-      # Check if port is listening using /dev/tcp
-      if timeout 1 bash -c "echo -n > /dev/tcp/localhost/$PROXY_PORT" 2>/dev/null; then
-        PROXY_READY=true
-        echo "✓ Proxy is ready on port $PROXY_PORT"
-        break
-      fi
-    else
-      echo "✗ FAILED: Error proxy process died"
-      break
-    fi
-    sleep 0.5
-  done
-
-  # Check if proxy is running and ready
-  if [ "$PROXY_READY" != "true" ]; then
-    echo "✗ FAILED: Error proxy failed to become ready"
-    echo "Proxy log:"
-    cat "$PROXY_LOG"
-    kill $PROXY_PID 2>/dev/null || true
-    RESULTS+=("✗ Out-of-Context Test Error (proxy failed)")
-  else
-    # Configure provider to use proxy and skip backoff
-    export ANTHROPIC_HOST="http://localhost:$PROXY_PORT"
-    export GOOSE_PROVIDER_SKIP_BACKOFF=true
-    export GOOSE_PROVIDER=anthropic
-    export GOOSE_MODEL=claude-haiku-4-5
-
-    echo "Step 1: Creating session (should trigger context-length error and compaction)..."
-    (cd "$TESTDIR" && "$GOOSE_BIN" run --text "hello world" 2>&1) | tee "$OUTPUT"
-
-    SESSION_ID=$("$GOOSE_BIN" session list --format json 2>/dev/null | jq -r '.[0].id' 2>/dev/null)
-
-    if [ -z "$SESSION_ID" ] || [ "$SESSION_ID" = "null" ]; then
-      echo "✗ FAILED: Could not create session"
-      RESULTS+=("✗ Out-of-Context Test Error (no session)")
-    else
-      echo ""
-      echo "Session created: $SESSION_ID"
-      echo "Checking for compaction evidence..."
-
-      # Check for compaction in the output
-      if grep -qi "context.*length\|compacting\|compacted\|compaction" "$OUTPUT"; then
-        echo "✓ SUCCESS: Out-of-context Test error triggered compaction"
-
-        if validate_compaction "$SESSION_ID" "out-of-context error compaction"; then
-          RESULTS+=("✓ Out-of-Context Test Error")
-        else
-          RESULTS+=("✗ Out-of-Context Test Error (structure validation failed)")
-        fi
-      else
-        echo "✗ FAILED: No evidence of compaction after context-length error"
-        echo "   Output:"
-        cat "$OUTPUT"
-        RESULTS+=("✗ Out-of-Context Test Error")
-      fi
-    fi
-
-    # Clean up
-    echo ""
-    echo "Stopping error proxy..."
-    # Kill the entire process group to ensure UV and Python processes are terminated
-    kill -- -$PROXY_PID 2>/dev/null || true
-    # Also explicitly kill any remaining UV processes on this port
-    pkill -f "uv run.*--port $PROXY_PORT" 2>/dev/null || true
-    wait $PROXY_PID 2>/dev/null || true
-    unset ANTHROPIC_HOST
-    unset GOOSE_PROVIDER_SKIP_BACKOFF
-    unset GOOSE_PROVIDER
-    unset GOOSE_MODEL
-    unset UV_INDEX_URL
-  fi
-fi
-
-rm -f "$OUTPUT" "$PROXY_LOG" "$PROXY_SETUP_LOG"
-rm -rf "$TESTDIR"
-
-echo ""
-echo ""
+# # ==================================================
+# # TEST 2: Auto Compaction
+# # ==================================================
+# echo "---------------------------------------------------"
+# echo "TEST 2: Auto Compaction via threshold (0.005)"
+# echo "---------------------------------------------------"
+
+# TESTDIR=$(mktemp -d)
+# echo "test content" > "$TESTDIR/test.txt"
+# echo "Test directory: $TESTDIR"
+# echo ""
+
+# # Set auto-compact threshold very low (.5%) to trigger it quickly
+# export GOOSE_AUTO_COMPACT_THRESHOLD=0.005
+
+# OUTPUT=$(mktemp)
+
+# LONG_RESPONSE_PROMPT="Count from 1 to 200, one number per line."
+
+# echo "Step 1: Creating session with first message (generating tokens for threshold)..."
+# (cd "$TESTDIR" && "$GOOSE_BIN" run --text "$LONG_RESPONSE_PROMPT" 2>&1) | tee "$OUTPUT"
+
+# if ! command -v jq &> /dev/null; then
+#   echo "✗ FAILED: jq is required for this test"
+#   RESULTS+=("✗ Auto Compaction (jq required)")
+# else
+#   SESSION_ID=$("$GOOSE_BIN" session list --format json 2>/dev/null | jq -r '.[0].id' 2>/dev/null)
+
+#   if [ -z "$SESSION_ID" ] || [ "$SESSION_ID" = "null" ]; then
+#     echo "✗ FAILED: Could not create session"
+#     RESULTS+=("✗ Auto Compaction (no session)")
+#   else
+#     echo ""
+#     echo "Session created: $SESSION_ID"
+#     echo "Step 2: Sending second message (should trigger auto-compact)..."
+
+#     # Send second message - auto-compaction should trigger before processing this
+#     (cd "$TESTDIR" && "$GOOSE_BIN" run --resume --session-id "$SESSION_ID" --text "hi again" 2>&1) | tee -a "$OUTPUT"
+
+#     echo ""
+#     echo "Checking for auto-compaction evidence..."
+
+#     if grep -qi "auto.*compact\|exceeded.*auto.*compact.*threshold" "$OUTPUT"; then
+#       echo "✓ SUCCESS: Auto compaction was triggered"
+
+#       if validate_compaction "$SESSION_ID" "auto compaction"; then
+#         RESULTS+=("✓ Auto Compaction")
+#       else
+#         RESULTS+=("✗ Auto Compaction (structure validation failed)")
+#       fi
+#     else
+#       echo "✗ FAILED: Auto compaction was not triggered"
+#       echo "   Expected to see auto-compact messages with threshold of 0.005"
+#       RESULTS+=("✗ Auto Compaction")
+#     fi
+#   fi
+# fi
+
+# # Unset the env variable
+# unset GOOSE_AUTO_COMPACT_THRESHOLD
+
+# rm -f "$OUTPUT"
+# rm -rf "$TESTDIR"
+
+# echo ""
+# echo ""
+
+# # ==================================================
+# # TEST 3: Out-of-Context Error Compaction
+# # ==================================================
+# echo "---------------------------------------------------"
+# echo "TEST 3: Compaction via out-of-context error (proxy)"
+# echo "---------------------------------------------------"
+
+# TESTDIR=$(mktemp -d)
+# echo "test content" > "$TESTDIR/test.txt"
+# echo "Test directory: $TESTDIR"
+# echo ""
+
+# # Use a random port to avoid conflicts
+# PROXY_PORT=$((9000 + RANDOM % 1000))
+# PROXY_DIR="$SCRIPT_DIR/scripts/provider-error-proxy"
+
+# OUTPUT=$(mktemp)
+# PROXY_LOG=$(mktemp)
+# PROXY_SETUP_LOG=$(mktemp)
+
+# # Pre-install proxy dependencies (so first run doesn't take forever)
+# echo "Installing proxy dependencies..."
+# export UV_INDEX_URL="https://pypi.org/simple"
+# if ! (cd "$PROXY_DIR" && uv sync 2>&1 | tee "$PROXY_SETUP_LOG"); then
+#   echo "✗ FAILED: Could not install proxy dependencies"
+#   echo "Setup log:"
+#   cat "$PROXY_SETUP_LOG"
+#   RESULTS+=("✗ Out-of-Context Error (dependency install failed)")
+# else
+#   echo "✓ Dependencies installed"
+
+#   # Start the error proxy in context-length error mode (3 errors)
+#   echo "Starting error proxy on port $PROXY_PORT with context-length error mode..."
+#   (cd "$PROXY_DIR" && UV_INDEX_URL="https://pypi.org/simple" uv run proxy.py --port "$PROXY_PORT" --mode "c 3" --no-stdin > "$PROXY_LOG" 2>&1) &
+#   PROXY_PID=$!
+
+#   # Wait for proxy to be ready (check if port is listening)
+#   echo "Waiting for proxy to be ready..."
+#   PROXY_READY=false
+#   for i in {1..60}; do
+#     if kill -0 $PROXY_PID 2>/dev/null; then
+#       # Check if port is listening using /dev/tcp
+#       if timeout 1 bash -c "echo -n > /dev/tcp/localhost/$PROXY_PORT" 2>/dev/null; then
+#         PROXY_READY=true
+#         echo "✓ Proxy is ready on port $PROXY_PORT"
+#         break
+#       fi
+#     else
+#       echo "✗ FAILED: Error proxy process died"
+#       break
+#     fi
+#     sleep 0.5
+#   done
+
+#   # Check if proxy is running and ready
+#   if [ "$PROXY_READY" != "true" ]; then
+#     echo "✗ FAILED: Error proxy failed to become ready"
+#     echo "Proxy log:"
+#     cat "$PROXY_LOG"
+#     kill $PROXY_PID 2>/dev/null || true
+#     RESULTS+=("✗ Out-of-Context Test Error (proxy failed)")
+#   else
+#     # Configure provider to use proxy and skip backoff
+#     export ANTHROPIC_HOST="http://localhost:$PROXY_PORT"
+#     export GOOSE_PROVIDER_SKIP_BACKOFF=true
+#     export GOOSE_PROVIDER=anthropic
+#     export GOOSE_MODEL=claude-haiku-4-5
+
+#     echo "Step 1: Creating session (should trigger context-length error and compaction)..."
+#     (cd "$TESTDIR" && "$GOOSE_BIN" run --text "hello world" 2>&1) | tee "$OUTPUT"
+
+#     SESSION_ID=$("$GOOSE_BIN" session list --format json 2>/dev/null | jq -r '.[0].id' 2>/dev/null)
+
+#     if [ -z "$SESSION_ID" ] || [ "$SESSION_ID" = "null" ]; then
+#       echo "✗ FAILED: Could not create session"
+#       RESULTS+=("✗ Out-of-Context Test Error (no session)")
+#     else
+#       echo ""
+#       echo "Session created: $SESSION_ID"
+#       echo "Checking for compaction evidence..."
+
+#       # Check for compaction in the output
+#       if grep -qi "context.*length\|compacting\|compacted\|compaction" "$OUTPUT"; then
+#         echo "✓ SUCCESS: Out-of-context Test error triggered compaction"
+
+#         if validate_compaction "$SESSION_ID" "out-of-context error compaction"; then
+#           RESULTS+=("✓ Out-of-Context Test Error")
+#         else
+#           RESULTS+=("✗ Out-of-Context Test Error (structure validation failed)")
+#         fi
+#       else
+#         echo "✗ FAILED: No evidence of compaction after context-length error"
+#         echo "   Output:"
+#         cat "$OUTPUT"
+#         RESULTS+=("✗ Out-of-Context Test Error")
+#       fi
+#     fi
+
+#     # Clean up
+#     echo ""
+#     echo "Stopping error proxy..."
+#     # Kill the entire process group to ensure UV and Python processes are terminated
+#     kill -- -$PROXY_PID 2>/dev/null || true
+#     # Also explicitly kill any remaining UV processes on this port
+#     pkill -f "uv run.*--port $PROXY_PORT" 2>/dev/null || true
+#     wait $PROXY_PID 2>/dev/null || true
+#     unset ANTHROPIC_HOST
+#     unset GOOSE_PROVIDER_SKIP_BACKOFF
+#     unset GOOSE_PROVIDER
+#     unset GOOSE_MODEL
+#     unset UV_INDEX_URL
+#   fi
+# fi
+
+# rm -f "$OUTPUT" "$PROXY_LOG" "$PROXY_SETUP_LOG"
+# rm -rf "$TESTDIR"
+
+# echo ""
+# echo ""
 
 # ==================================================
 # Summary

From 512b5f6a62505bbe838290cdd9115b754f1f514f Mon Sep 17 00:00:00 2001
From: Lifei Zhou <lifei@squareup.com>
Date: Tue, 27 Jan 2026 23:30:50 +1100
Subject: [PATCH 2/7] use correct logfile

---
 scripts/test_compaction.sh | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/scripts/test_compaction.sh b/scripts/test_compaction.sh
index ef95556bcb3f..8115a01673bb 100755
--- a/scripts/test_compaction.sh
+++ b/scripts/test_compaction.sh
@@ -22,8 +22,19 @@ GOOSE_BIN="$SCRIPT_DIR/target/release/goose"
 # Timeout for goose commands (default 5 minutes)
 GOOSE_TIMEOUT=${GOOSE_TIMEOUT:-300}
 
-# Log directory for diagnostics
-LOG_DIR="$HOME/.config/goose/logs/cli"
+# Log directory for diagnostics - must match goose's actual log path
+# See crates/goose/src/config/paths.rs and crates/goose/src/logging.rs
+get_log_dir() {
+  local today=$(date +%Y-%m-%d)
+  if [ -n "$GOOSE_PATH_ROOT" ]; then
+    echo "$GOOSE_PATH_ROOT/state/logs/cli/$today"
+  elif [ "$(uname)" = "Darwin" ]; then
+    echo "$HOME/Library/Application Support/Block.goose/logs/cli/$today"
+  else
+    echo "${XDG_STATE_HOME:-$HOME/.local/state}/goose/logs/cli/$today"
+  fi
+}
+LOG_DIR=$(get_log_dir)
 
 # Helper to dump stream diagnostics from logs on timeout
 dump_stream_diagnostics() {
@@ -31,7 +42,7 @@ dump_stream_diagnostics() {
   echo "=== STREAM DIAGNOSTICS (last 50 entries) ==="
   if [ -d "$LOG_DIR" ]; then
     # Find logs from the last 10 minutes and grep for stream_diag
-    find "$LOG_DIR" -name "*.log" -mmin -10 -exec grep -h "stream_diag" {} \; 2>/dev/null | tail -50
+    find "$LOG_DIR" -name "*.log" -mmin -10 -exec grep -h "stream_diag" {} \; 2>/dev/null | tail -200
   else
     echo "No log directory found at $LOG_DIR"
   fi

From ae1ade599e5070f82d1b4052db748113f6fb4950 Mon Sep 17 00:00:00 2001
From: Lifei Zhou <lifei@squareup.com>
Date: Wed, 28 Jan 2026 01:36:08 +1100
Subject: [PATCH 3/7] Revert "use correct logfile"

This reverts commit 512b5f6a62505bbe838290cdd9115b754f1f514f.
---
 scripts/test_compaction.sh | 17 +++--------------
 1 file changed, 3 insertions(+), 14 deletions(-)

diff --git a/scripts/test_compaction.sh b/scripts/test_compaction.sh
index 8115a01673bb..ef95556bcb3f 100755
--- a/scripts/test_compaction.sh
+++ b/scripts/test_compaction.sh
@@ -22,19 +22,8 @@ GOOSE_BIN="$SCRIPT_DIR/target/release/goose"
 # Timeout for goose commands (default 5 minutes)
 GOOSE_TIMEOUT=${GOOSE_TIMEOUT:-300}
 
-# Log directory for diagnostics - must match goose's actual log path
-# See crates/goose/src/config/paths.rs and crates/goose/src/logging.rs
-get_log_dir() {
-  local today=$(date +%Y-%m-%d)
-  if [ -n "$GOOSE_PATH_ROOT" ]; then
-    echo "$GOOSE_PATH_ROOT/state/logs/cli/$today"
-  elif [ "$(uname)" = "Darwin" ]; then
-    echo "$HOME/Library/Application Support/Block.goose/logs/cli/$today"
-  else
-    echo "${XDG_STATE_HOME:-$HOME/.local/state}/goose/logs/cli/$today"
-  fi
-}
-LOG_DIR=$(get_log_dir)
+# Log directory for diagnostics
+LOG_DIR="$HOME/.config/goose/logs/cli"
 
 # Helper to dump stream diagnostics from logs on timeout
 dump_stream_diagnostics() {
@@ -42,7 +31,7 @@ dump_stream_diagnostics() {
   echo "=== STREAM DIAGNOSTICS (last 50 entries) ==="
   if [ -d "$LOG_DIR" ]; then
     # Find logs from the last 10 minutes and grep for stream_diag
-    find "$LOG_DIR" -name "*.log" -mmin -10 -exec grep -h "stream_diag" {} \; 2>/dev/null | tail -200
+    find "$LOG_DIR" -name "*.log" -mmin -10 -exec grep -h "stream_diag" {} \; 2>/dev/null | tail -50
   else
     echo "No log directory found at $LOG_DIR"
   fi

From 41338e311ad39a05fc07486df2e27dd1a913ec92 Mon Sep 17 00:00:00 2001
From: Lifei Zhou <lifei@squareup.com>
Date: Wed, 28 Jan 2026 01:36:58 +1100
Subject: [PATCH 4/7] Revert "debug"

This reverts commit afbd04b5d62a890984764cc67efcb2c4ff67b4fe.
---
 crates/goose/src/agents/subagent_handler.rs   |  18 -
 .../goose/src/providers/formats/anthropic.rs  |  16 -
 scripts/test_compaction.sh                    | 413 ++++++++----------
 3 files changed, 188 insertions(+), 259 deletions(-)

diff --git a/crates/goose/src/agents/subagent_handler.rs b/crates/goose/src/agents/subagent_handler.rs
index c2683421894d..54e0d2dab827 100644
--- a/crates/goose/src/agents/subagent_handler.rs
+++ b/crates/goose/src/agents/subagent_handler.rs
@@ -186,22 +186,7 @@ fn get_agent_messages(
             .reply(user_message, session_config, cancellation_token)
             .await
             .map_err(|e| anyhow!("Failed to get reply from agent: {}", e))?;
-
-        // Stream diagnostics for debugging hangs
-        let mut event_count = 0u64;
-        let stream_start = std::time::Instant::now();
-        let mut last_event_time = stream_start;
-        tracing::info!(target: "stream_diag", "[STREAM] subagent: starting to consume agent events");
-
         while let Some(message_result) = stream.next().await {
-            event_count += 1;
-            let now = std::time::Instant::now();
-            let since_last = now.duration_since(last_event_time);
-            if since_last.as_secs() >= 5 {
-                tracing::warn!(target: "stream_diag", "[STREAM] subagent: event #{} after {:.1}s delay", event_count, since_last.as_secs_f64());
-            }
-            last_event_time = now;
-
             match message_result {
                 Ok(AgentEvent::Message(msg)) => conversation.push(msg),
                 Ok(AgentEvent::McpNotification(_)) | Ok(AgentEvent::ModelChange { .. }) => {}
@@ -215,9 +200,6 @@ fn get_agent_messages(
             }
         }
 
-        let total_duration = stream_start.elapsed();
-        tracing::info!(target: "stream_diag", "[STREAM] subagent: stream ended after {} events in {:.1}s", event_count, total_duration.as_secs_f64());
-
         let final_output = if has_response_schema {
             agent
                 .final_output_tool
diff --git a/crates/goose/src/providers/formats/anthropic.rs b/crates/goose/src/providers/formats/anthropic.rs
index 61d6622563ff..7e3dd161515a 100644
--- a/crates/goose/src/providers/formats/anthropic.rs
+++ b/crates/goose/src/providers/formats/anthropic.rs
@@ -494,20 +494,7 @@ where
         let mut final_usage: Option<crate::providers::base::ProviderUsage> = None;
         let mut message_id: Option<String> = None;
 
-        // Stream diagnostics for debugging hangs
-        let mut chunk_count = 0u64;
-        let stream_start = std::time::Instant::now();
-        let mut last_chunk_time = stream_start;
-        tracing::info!(target: "stream_diag", "[STREAM] anthropic: starting stream consumption");
-
         while let Some(line_result) = stream.next().await {
-            chunk_count += 1;
-            let now = std::time::Instant::now();
-            let since_last = now.duration_since(last_chunk_time);
-            if since_last.as_secs() >= 5 {
-                tracing::warn!(target: "stream_diag", "[STREAM] anthropic: chunk #{} after {:.1}s delay", chunk_count, since_last.as_secs_f64());
-            }
-            last_chunk_time = now;
             let line = line_result?;
 
             // Skip empty lines and non-data lines
@@ -709,9 +696,6 @@ where
             }
         }
 
-        let total_duration = stream_start.elapsed();
-        tracing::info!(target: "stream_diag", "[STREAM] anthropic: stream ended after {} chunks in {:.1}s", chunk_count, total_duration.as_secs_f64());
-
         // Yield final usage information if available
         if let Some(usage) = final_usage {
             yield (None, Some(usage));
diff --git a/scripts/test_compaction.sh b/scripts/test_compaction.sh
index ef95556bcb3f..baa3b7ca9660 100755
--- a/scripts/test_compaction.sh
+++ b/scripts/test_compaction.sh
@@ -19,43 +19,6 @@ fi
 SCRIPT_DIR=$(pwd)
 GOOSE_BIN="$SCRIPT_DIR/target/release/goose"
 
-# Timeout for goose commands (default 5 minutes)
-GOOSE_TIMEOUT=${GOOSE_TIMEOUT:-300}
-
-# Log directory for diagnostics
-LOG_DIR="$HOME/.config/goose/logs/cli"
-
-# Helper to dump stream diagnostics from logs on timeout
-dump_stream_diagnostics() {
-  echo ""
-  echo "=== STREAM DIAGNOSTICS (last 50 entries) ==="
-  if [ -d "$LOG_DIR" ]; then
-    # Find logs from the last 10 minutes and grep for stream_diag
-    find "$LOG_DIR" -name "*.log" -mmin -10 -exec grep -h "stream_diag" {} \; 2>/dev/null | tail -50
-  else
-    echo "No log directory found at $LOG_DIR"
-  fi
-  echo "=== END STREAM DIAGNOSTICS ==="
-  echo ""
-}
-
-# Helper to run goose with timeout and stream diagnostics enabled
-run_goose() {
-  local exit_code=0
-  # Enable stream_diag target for our diagnostic logs
-  RUST_LOG="${RUST_LOG:-stream_diag=info}" timeout "$GOOSE_TIMEOUT" "$GOOSE_BIN" "$@" 2>&1 || exit_code=$?
-
-  if [ $exit_code -eq 124 ] || [ $exit_code -eq 137 ]; then
-    echo ""
-    echo "⚠️  TIMEOUT: goose command exceeded ${GOOSE_TIMEOUT}s limit"
-  fi
-
-  # Always dump stream diagnostics to help debug issues
-  dump_stream_diagnostics
-
-  return $exit_code
-}
-
 # Validation function to check compaction structure in session JSON
 validate_compaction() {
   local session_id=$1
@@ -139,7 +102,7 @@ echo ""
 OUTPUT=$(mktemp)
 
 echo "Step 1: Creating session with initial messages..."
-(cd "$TESTDIR" && run_goose run --text "list files and read hello.txt") | tee "$OUTPUT"
+(cd "$TESTDIR" && "$GOOSE_BIN" run --text "list files and read hello.txt" 2>&1) | tee "$OUTPUT"
 
 if ! command -v jq &> /dev/null; then
   echo "✗ FAILED: jq is required for this test"
@@ -158,7 +121,7 @@ else
     echo "Step 2: Sending manual compaction trigger..."
 
     # Send the manual compact trigger prompt
-    (cd "$TESTDIR" && run_goose run --resume --session-id "$SESSION_ID" --text "Please compact this conversation") | tee -a "$OUTPUT"
+    (cd "$TESTDIR" && "$GOOSE_BIN" run --resume --session-id "$SESSION_ID" --text "Please compact this conversation" 2>&1) | tee -a "$OUTPUT"
 
     echo ""
     echo "Checking for compaction evidence..."
@@ -184,192 +147,192 @@ fi
 echo ""
 echo ""
 
-# # ==================================================
-# # TEST 2: Auto Compaction
-# # ==================================================
-# echo "---------------------------------------------------"
-# echo "TEST 2: Auto Compaction via threshold (0.005)"
-# echo "---------------------------------------------------"
-
-# TESTDIR=$(mktemp -d)
-# echo "test content" > "$TESTDIR/test.txt"
-# echo "Test directory: $TESTDIR"
-# echo ""
-
-# # Set auto-compact threshold very low (.5%) to trigger it quickly
-# export GOOSE_AUTO_COMPACT_THRESHOLD=0.005
-
-# OUTPUT=$(mktemp)
-
-# LONG_RESPONSE_PROMPT="Count from 1 to 200, one number per line."
-
-# echo "Step 1: Creating session with first message (generating tokens for threshold)..."
-# (cd "$TESTDIR" && "$GOOSE_BIN" run --text "$LONG_RESPONSE_PROMPT" 2>&1) | tee "$OUTPUT"
-
-# if ! command -v jq &> /dev/null; then
-#   echo "✗ FAILED: jq is required for this test"
-#   RESULTS+=("✗ Auto Compaction (jq required)")
-# else
-#   SESSION_ID=$("$GOOSE_BIN" session list --format json 2>/dev/null | jq -r '.[0].id' 2>/dev/null)
-
-#   if [ -z "$SESSION_ID" ] || [ "$SESSION_ID" = "null" ]; then
-#     echo "✗ FAILED: Could not create session"
-#     RESULTS+=("✗ Auto Compaction (no session)")
-#   else
-#     echo ""
-#     echo "Session created: $SESSION_ID"
-#     echo "Step 2: Sending second message (should trigger auto-compact)..."
-
-#     # Send second message - auto-compaction should trigger before processing this
-#     (cd "$TESTDIR" && "$GOOSE_BIN" run --resume --session-id "$SESSION_ID" --text "hi again" 2>&1) | tee -a "$OUTPUT"
-
-#     echo ""
-#     echo "Checking for auto-compaction evidence..."
-
-#     if grep -qi "auto.*compact\|exceeded.*auto.*compact.*threshold" "$OUTPUT"; then
-#       echo "✓ SUCCESS: Auto compaction was triggered"
-
-#       if validate_compaction "$SESSION_ID" "auto compaction"; then
-#         RESULTS+=("✓ Auto Compaction")
-#       else
-#         RESULTS+=("✗ Auto Compaction (structure validation failed)")
-#       fi
-#     else
-#       echo "✗ FAILED: Auto compaction was not triggered"
-#       echo "   Expected to see auto-compact messages with threshold of 0.005"
-#       RESULTS+=("✗ Auto Compaction")
-#     fi
-#   fi
-# fi
-
-# # Unset the env variable
-# unset GOOSE_AUTO_COMPACT_THRESHOLD
-
-# rm -f "$OUTPUT"
-# rm -rf "$TESTDIR"
-
-# echo ""
-# echo ""
-
-# # ==================================================
-# # TEST 3: Out-of-Context Error Compaction
-# # ==================================================
-# echo "---------------------------------------------------"
-# echo "TEST 3: Compaction via out-of-context error (proxy)"
-# echo "---------------------------------------------------"
-
-# TESTDIR=$(mktemp -d)
-# echo "test content" > "$TESTDIR/test.txt"
-# echo "Test directory: $TESTDIR"
-# echo ""
-
-# # Use a random port to avoid conflicts
-# PROXY_PORT=$((9000 + RANDOM % 1000))
-# PROXY_DIR="$SCRIPT_DIR/scripts/provider-error-proxy"
-
-# OUTPUT=$(mktemp)
-# PROXY_LOG=$(mktemp)
-# PROXY_SETUP_LOG=$(mktemp)
-
-# # Pre-install proxy dependencies (so first run doesn't take forever)
-# echo "Installing proxy dependencies..."
-# export UV_INDEX_URL="https://pypi.org/simple"
-# if ! (cd "$PROXY_DIR" && uv sync 2>&1 | tee "$PROXY_SETUP_LOG"); then
-#   echo "✗ FAILED: Could not install proxy dependencies"
-#   echo "Setup log:"
-#   cat "$PROXY_SETUP_LOG"
-#   RESULTS+=("✗ Out-of-Context Error (dependency install failed)")
-# else
-#   echo "✓ Dependencies installed"
-
-#   # Start the error proxy in context-length error mode (3 errors)
-#   echo "Starting error proxy on port $PROXY_PORT with context-length error mode..."
-#   (cd "$PROXY_DIR" && UV_INDEX_URL="https://pypi.org/simple" uv run proxy.py --port "$PROXY_PORT" --mode "c 3" --no-stdin > "$PROXY_LOG" 2>&1) &
-#   PROXY_PID=$!
-
-#   # Wait for proxy to be ready (check if port is listening)
-#   echo "Waiting for proxy to be ready..."
-#   PROXY_READY=false
-#   for i in {1..60}; do
-#     if kill -0 $PROXY_PID 2>/dev/null; then
-#       # Check if port is listening using /dev/tcp
-#       if timeout 1 bash -c "echo -n > /dev/tcp/localhost/$PROXY_PORT" 2>/dev/null; then
-#         PROXY_READY=true
-#         echo "✓ Proxy is ready on port $PROXY_PORT"
-#         break
-#       fi
-#     else
-#       echo "✗ FAILED: Error proxy process died"
-#       break
-#     fi
-#     sleep 0.5
-#   done
-
-#   # Check if proxy is running and ready
-#   if [ "$PROXY_READY" != "true" ]; then
-#     echo "✗ FAILED: Error proxy failed to become ready"
-#     echo "Proxy log:"
-#     cat "$PROXY_LOG"
-#     kill $PROXY_PID 2>/dev/null || true
-#     RESULTS+=("✗ Out-of-Context Test Error (proxy failed)")
-#   else
-#     # Configure provider to use proxy and skip backoff
-#     export ANTHROPIC_HOST="http://localhost:$PROXY_PORT"
-#     export GOOSE_PROVIDER_SKIP_BACKOFF=true
-#     export GOOSE_PROVIDER=anthropic
-#     export GOOSE_MODEL=claude-haiku-4-5
-
-#     echo "Step 1: Creating session (should trigger context-length error and compaction)..."
-#     (cd "$TESTDIR" && "$GOOSE_BIN" run --text "hello world" 2>&1) | tee "$OUTPUT"
-
-#     SESSION_ID=$("$GOOSE_BIN" session list --format json 2>/dev/null | jq -r '.[0].id' 2>/dev/null)
-
-#     if [ -z "$SESSION_ID" ] || [ "$SESSION_ID" = "null" ]; then
-#       echo "✗ FAILED: Could not create session"
-#       RESULTS+=("✗ Out-of-Context Test Error (no session)")
-#     else
-#       echo ""
-#       echo "Session created: $SESSION_ID"
-#       echo "Checking for compaction evidence..."
-
-#       # Check for compaction in the output
-#       if grep -qi "context.*length\|compacting\|compacted\|compaction" "$OUTPUT"; then
-#         echo "✓ SUCCESS: Out-of-context Test error triggered compaction"
-
-#         if validate_compaction "$SESSION_ID" "out-of-context error compaction"; then
-#           RESULTS+=("✓ Out-of-Context Test Error")
-#         else
-#           RESULTS+=("✗ Out-of-Context Test Error (structure validation failed)")
-#         fi
-#       else
-#         echo "✗ FAILED: No evidence of compaction after context-length error"
-#         echo "   Output:"
-#         cat "$OUTPUT"
-#         RESULTS+=("✗ Out-of-Context Test Error")
-#       fi
-#     fi
-
-#     # Clean up
-#     echo ""
-#     echo "Stopping error proxy..."
-#     # Kill the entire process group to ensure UV and Python processes are terminated
-#     kill -- -$PROXY_PID 2>/dev/null || true
-#     # Also explicitly kill any remaining UV processes on this port
-#     pkill -f "uv run.*--port $PROXY_PORT" 2>/dev/null || true
-#     wait $PROXY_PID 2>/dev/null || true
-#     unset ANTHROPIC_HOST
-#     unset GOOSE_PROVIDER_SKIP_BACKOFF
-#     unset GOOSE_PROVIDER
-#     unset GOOSE_MODEL
-#     unset UV_INDEX_URL
-#   fi
-# fi
-
-# rm -f "$OUTPUT" "$PROXY_LOG" "$PROXY_SETUP_LOG"
-# rm -rf "$TESTDIR"
-
-# echo ""
-# echo ""
+# ==================================================
+# TEST 2: Auto Compaction
+# ==================================================
+echo "---------------------------------------------------"
+echo "TEST 2: Auto Compaction via threshold (0.005)"
+echo "---------------------------------------------------"
+
+TESTDIR=$(mktemp -d)
+echo "test content" > "$TESTDIR/test.txt"
+echo "Test directory: $TESTDIR"
+echo ""
+
+# Set auto-compact threshold very low (.5%) to trigger it quickly
+export GOOSE_AUTO_COMPACT_THRESHOLD=0.005
+
+OUTPUT=$(mktemp)
+
+LONG_RESPONSE_PROMPT="Count from 1 to 200, one number per line."
+
+echo "Step 1: Creating session with first message (generating tokens for threshold)..."
+(cd "$TESTDIR" && "$GOOSE_BIN" run --text "$LONG_RESPONSE_PROMPT" 2>&1) | tee "$OUTPUT"
+
+if ! command -v jq &> /dev/null; then
+  echo "✗ FAILED: jq is required for this test"
+  RESULTS+=("✗ Auto Compaction (jq required)")
+else
+  SESSION_ID=$("$GOOSE_BIN" session list --format json 2>/dev/null | jq -r '.[0].id' 2>/dev/null)
+
+  if [ -z "$SESSION_ID" ] || [ "$SESSION_ID" = "null" ]; then
+    echo "✗ FAILED: Could not create session"
+    RESULTS+=("✗ Auto Compaction (no session)")
+  else
+    echo ""
+    echo "Session created: $SESSION_ID"
+    echo "Step 2: Sending second message (should trigger auto-compact)..."
+
+    # Send second message - auto-compaction should trigger before processing this
+    (cd "$TESTDIR" && "$GOOSE_BIN" run --resume --session-id "$SESSION_ID" --text "hi again" 2>&1) | tee -a "$OUTPUT"
+
+    echo ""
+    echo "Checking for auto-compaction evidence..."
+
+    if grep -qi "auto.*compact\|exceeded.*auto.*compact.*threshold" "$OUTPUT"; then
+      echo "✓ SUCCESS: Auto compaction was triggered"
+
+      if validate_compaction "$SESSION_ID" "auto compaction"; then
+        RESULTS+=("✓ Auto Compaction")
+      else
+        RESULTS+=("✗ Auto Compaction (structure validation failed)")
+      fi
+    else
+      echo "✗ FAILED: Auto compaction was not triggered"
+      echo "   Expected to see auto-compact messages with threshold of 0.005"
+      RESULTS+=("✗ Auto Compaction")
+    fi
+  fi
+fi
+
+# Unset the env variable
+unset GOOSE_AUTO_COMPACT_THRESHOLD
+
+rm -f "$OUTPUT"
+rm -rf "$TESTDIR"
+
+echo ""
+echo ""
+
+# ==================================================
+# TEST 3: Out-of-Context Error Compaction
+# ==================================================
+echo "---------------------------------------------------"
+echo "TEST 3: Compaction via out-of-context error (proxy)"
+echo "---------------------------------------------------"
+
+TESTDIR=$(mktemp -d)
+echo "test content" > "$TESTDIR/test.txt"
+echo "Test directory: $TESTDIR"
+echo ""
+
+# Use a random port to avoid conflicts
+PROXY_PORT=$((9000 + RANDOM % 1000))
+PROXY_DIR="$SCRIPT_DIR/scripts/provider-error-proxy"
+
+OUTPUT=$(mktemp)
+PROXY_LOG=$(mktemp)
+PROXY_SETUP_LOG=$(mktemp)
+
+# Pre-install proxy dependencies (so first run doesn't take forever)
+echo "Installing proxy dependencies..."
+export UV_INDEX_URL="https://pypi.org/simple"
+if ! (cd "$PROXY_DIR" && uv sync 2>&1 | tee "$PROXY_SETUP_LOG"); then
+  echo "✗ FAILED: Could not install proxy dependencies"
+  echo "Setup log:"
+  cat "$PROXY_SETUP_LOG"
+  RESULTS+=("✗ Out-of-Context Error (dependency install failed)")
+else
+  echo "✓ Dependencies installed"
+
+  # Start the error proxy in context-length error mode (3 errors)
+  echo "Starting error proxy on port $PROXY_PORT with context-length error mode..."
+  (cd "$PROXY_DIR" && UV_INDEX_URL="https://pypi.org/simple" uv run proxy.py --port "$PROXY_PORT" --mode "c 3" --no-stdin > "$PROXY_LOG" 2>&1) &
+  PROXY_PID=$!
+
+  # Wait for proxy to be ready (check if port is listening)
+  echo "Waiting for proxy to be ready..."
+  PROXY_READY=false
+  for i in {1..60}; do
+    if kill -0 $PROXY_PID 2>/dev/null; then
+      # Check if port is listening using /dev/tcp
+      if timeout 1 bash -c "echo -n > /dev/tcp/localhost/$PROXY_PORT" 2>/dev/null; then
+        PROXY_READY=true
+        echo "✓ Proxy is ready on port $PROXY_PORT"
+        break
+      fi
+    else
+      echo "✗ FAILED: Error proxy process died"
+      break
+    fi
+    sleep 0.5
+  done
+
+  # Check if proxy is running and ready
+  if [ "$PROXY_READY" != "true" ]; then
+    echo "✗ FAILED: Error proxy failed to become ready"
+    echo "Proxy log:"
+    cat "$PROXY_LOG"
+    kill $PROXY_PID 2>/dev/null || true
+    RESULTS+=("✗ Out-of-Context Test Error (proxy failed)")
+  else
+    # Configure provider to use proxy and skip backoff
+    export ANTHROPIC_HOST="http://localhost:$PROXY_PORT"
+    export GOOSE_PROVIDER_SKIP_BACKOFF=true
+    export GOOSE_PROVIDER=anthropic
+    export GOOSE_MODEL=claude-haiku-4-5
+
+    echo "Step 1: Creating session (should trigger context-length error and compaction)..."
+    (cd "$TESTDIR" && "$GOOSE_BIN" run --text "hello world" 2>&1) | tee "$OUTPUT"
+
+    SESSION_ID=$("$GOOSE_BIN" session list --format json 2>/dev/null | jq -r '.[0].id' 2>/dev/null)
+
+    if [ -z "$SESSION_ID" ] || [ "$SESSION_ID" = "null" ]; then
+      echo "✗ FAILED: Could not create session"
+      RESULTS+=("✗ Out-of-Context Test Error (no session)")
+    else
+      echo ""
+      echo "Session created: $SESSION_ID"
+      echo "Checking for compaction evidence..."
+
+      # Check for compaction in the output
+      if grep -qi "context.*length\|compacting\|compacted\|compaction" "$OUTPUT"; then
+        echo "✓ SUCCESS: Out-of-context Test error triggered compaction"
+
+        if validate_compaction "$SESSION_ID" "out-of-context error compaction"; then
+          RESULTS+=("✓ Out-of-Context Test Error")
+        else
+          RESULTS+=("✗ Out-of-Context Test Error (structure validation failed)")
+        fi
+      else
+        echo "✗ FAILED: No evidence of compaction after context-length error"
+        echo "   Output:"
+        cat "$OUTPUT"
+        RESULTS+=("✗ Out-of-Context Test Error")
+      fi
+    fi
+
+    # Clean up
+    echo ""
+    echo "Stopping error proxy..."
+    # Kill the entire process group to ensure UV and Python processes are terminated
+    kill -- -$PROXY_PID 2>/dev/null || true
+    # Also explicitly kill any remaining UV processes on this port
+    pkill -f "uv run.*--port $PROXY_PORT" 2>/dev/null || true
+    wait $PROXY_PID 2>/dev/null || true
+    unset ANTHROPIC_HOST
+    unset GOOSE_PROVIDER_SKIP_BACKOFF
+    unset GOOSE_PROVIDER
+    unset GOOSE_MODEL
+    unset UV_INDEX_URL
+  fi
+fi
+
+rm -f "$OUTPUT" "$PROXY_LOG" "$PROXY_SETUP_LOG"
+rm -rf "$TESTDIR"
+
+echo ""
+echo ""
 
 # ==================================================
 # Summary

From e54cd999c537f9f2e5c8e2e6fa469c76db9fcf0c Mon Sep 17 00:00:00 2001
From: Lifei Zhou <lifei@squareup.com>
Date: Wed, 28 Jan 2026 01:43:06 +1100
Subject: [PATCH 5/7] fix it with a prompt that does not depend on developer
 tool

---
 scripts/test_compaction.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/test_compaction.sh b/scripts/test_compaction.sh
index baa3b7ca9660..248dfbde3455 100755
--- a/scripts/test_compaction.sh
+++ b/scripts/test_compaction.sh
@@ -102,7 +102,7 @@ echo ""
 OUTPUT=$(mktemp)
 
 echo "Step 1: Creating session with initial messages..."
-(cd "$TESTDIR" && "$GOOSE_BIN" run --text "list files and read hello.txt" 2>&1) | tee "$OUTPUT"
+(cd "$TESTDIR" && "$GOOSE_BIN" run --text "describe Melbourne Australia in 100 words" 2>&1) | tee "$OUTPUT"
 
 if ! command -v jq &> /dev/null; then
   echo "✗ FAILED: jq is required for this test"

From 136fab05f810dc79cc5f757bc6720788bb25791e Mon Sep 17 00:00:00 2001
From: Lifei Zhou <lifei@squareup.com>
Date: Wed, 28 Jan 2026 16:32:30 +1100
Subject: [PATCH 6/7] fixed the deadlock when read_resource tool call

---
 crates/goose/src/agents/extension_manager.rs | 11 ++++++++---
 crates/goose/src/prompts/system.md           |  4 ++--
 scripts/test_compaction.sh                   |  2 +-
 3 files changed, 11 insertions(+), 6 deletions(-)

diff --git a/crates/goose/src/agents/extension_manager.rs b/crates/goose/src/agents/extension_manager.rs
index 13869241f09b..995cbdc84b90 100644
--- a/crates/goose/src/agents/extension_manager.rs
+++ b/crates/goose/src/agents/extension_manager.rs
@@ -913,9 +913,14 @@ impl ExtensionManager {
         // Loop through each extension and try to read the resource, don't raise an error if the resource is not found
         // TODO: do we want to find if a provided uri is in multiple extensions?
         // currently it will return the first match and skip any others
-
-        // Collect extension names first to avoid holding the lock during iteration
-        let extension_names: Vec<String> = self.extensions.lock().await.keys().cloned().collect();
+        let extension_names: Vec<String> = self
+            .extensions
+            .lock()
+            .await
+            .iter()
+            .filter(|(_name, ext)| ext.supports_resources())
+            .map(|(name, _)| name.clone())
+            .collect();
 
         for extension_name in extension_names {
             let read_result = self
diff --git a/crates/goose/src/prompts/system.md b/crates/goose/src/prompts/system.md
index 87093d3997b8..c282d75497ac 100644
--- a/crates/goose/src/prompts/system.md
+++ b/crates/goose/src/prompts/system.md
@@ -31,8 +31,8 @@ in your tool specification.
 ## {{extension.name}}
 
 {% if extension.has_resources %}
-{{extension.name}} supports resources, you can use platform__read_resource,
-and platform__list_resources on this extension.
+{{extension.name}} supports resources, you can use extensionmanager__read_resource,
+and extensionmanager__list_resources on this extension.
 {% endif %}
 {% if extension.instructions %}### Instructions
 {{extension.instructions}}{% endif %}
diff --git a/scripts/test_compaction.sh b/scripts/test_compaction.sh
index 248dfbde3455..baa3b7ca9660 100755
--- a/scripts/test_compaction.sh
+++ b/scripts/test_compaction.sh
@@ -102,7 +102,7 @@ echo ""
 OUTPUT=$(mktemp)
 
 echo "Step 1: Creating session with initial messages..."
-(cd "$TESTDIR" && "$GOOSE_BIN" run --text "describe Melbourne Australia in 100 words" 2>&1) | tee "$OUTPUT"
+(cd "$TESTDIR" && "$GOOSE_BIN" run --text "list files and read hello.txt" 2>&1) | tee "$OUTPUT"
 
 if ! command -v jq &> /dev/null; then
   echo "✗ FAILED: jq is required for this test"

From 51e38448df7c16846b3261419e54ad99119032d4 Mon Sep 17 00:00:00 2001
From: Lifei Zhou <lifei@squareup.com>
Date: Wed, 28 Jan 2026 16:55:40 +1100
Subject: [PATCH 7/7] fixed the test

---
 .../goose__agents__prompt_manager__tests__one_extension.snap  | 4 ++--
 .../goose__agents__prompt_manager__tests__typical_setup.snap  | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/crates/goose/src/agents/snapshots/goose__agents__prompt_manager__tests__one_extension.snap b/crates/goose/src/agents/snapshots/goose__agents__prompt_manager__tests__one_extension.snap
index d7ba6de0dfec..8f02d1cf7af1 100644
--- a/crates/goose/src/agents/snapshots/goose__agents__prompt_manager__tests__one_extension.snap
+++ b/crates/goose/src/agents/snapshots/goose__agents__prompt_manager__tests__one_extension.snap
@@ -31,8 +31,8 @@ in your tool specification.
 
 ## test
 
-test supports resources, you can use platform__read_resource,
-and platform__list_resources on this extension.
+test supports resources, you can use extensionmanager__read_resource,
+and extensionmanager__list_resources on this extension.
 ### Instructions
 how to use this extension
 
diff --git a/crates/goose/src/agents/snapshots/goose__agents__prompt_manager__tests__typical_setup.snap b/crates/goose/src/agents/snapshots/goose__agents__prompt_manager__tests__typical_setup.snap
index c9848fd9ca20..719a84871da8 100644
--- a/crates/goose/src/agents/snapshots/goose__agents__prompt_manager__tests__typical_setup.snap
+++ b/crates/goose/src/agents/snapshots/goose__agents__prompt_manager__tests__typical_setup.snap
@@ -31,8 +31,8 @@ in your tool specification.
 
 ## extension_A
 
-extension_A supports resources, you can use platform__read_resource,
-and platform__list_resources on this extension.
+extension_A supports resources, you can use extensionmanager__read_resource,
+and extensionmanager__list_resources on this extension.
 ### Instructions
 <instructions on how to use extension A>
 ## extension_B