loki-mode 6.71.1 → 6.72.0

package/autonomy/run.sh

@@ -947,8 +947,9 @@ emit_event_json() {
         else
             json_data+=","
         fi
-        # Quote string values, leave numbers/booleans as-is
-        if [[ "$value" =~ ^[0-9]+$ ]] || [[ "$value" =~ ^(true|false|null)$ ]]; then
+        # Quote string values, leave numbers/booleans/floats as-is
+        # BUG-NEW-004: Also match floats (e.g., cost=3.14) not just integers
+        if [[ "$value" =~ ^[0-9]+\.?[0-9]*$ ]] || [[ "$value" =~ ^(true|false|null)$ ]]; then
             json_data+="\"$key\":$value"
         else
             # Escape backslashes, quotes, and special chars in value
@@ -1364,6 +1365,9 @@ get_phase_names() {
 
 # Global tier for current iteration (set by get_rarv_tier)
 CURRENT_TIER="development"
+# Export for provider helper functions (e.g., gemini.sh:provider_get_current_model)
+LOKI_CURRENT_TIER="$CURRENT_TIER"
+export LOKI_CURRENT_TIER
 
 # Get the appropriate tier based on RARV cycle step
 # Args: iteration_count (defaults to ITERATION_COUNT)
@@ -2159,6 +2163,17 @@ create_worktree() {
         return 0
     else
         log_error "Failed to create worktree: $stream_name"
+        # BUG-PU-001: Clean up partial worktree on creation failure
+        if [ -d "$worktree_path" ]; then
+            git -C "$TARGET_DIR" worktree remove "$worktree_path" --force 2>/dev/null || \
+                rm -rf "$worktree_path" 2>/dev/null || true
+        fi
+        # Clean up any orphaned branch created during the attempt
+        if [ -n "$branch_name" ]; then
+            git -C "$TARGET_DIR" branch -D "$branch_name" 2>/dev/null || true
+        else
+            git -C "$TARGET_DIR" branch -D "parallel-${stream_name}" 2>/dev/null || true
+        fi
         return 1
     fi
 }
@@ -3026,7 +3041,15 @@ invoke_gemini() {
     local prompt="$1"
     shift
 
-    local model="${PROVIDER_MODEL:-${GEMINI_DEFAULT_PRO:-gemini-3-pro-preview}}"
+    # BUG-PROV-001/006 fix: Use dynamic model resolution instead of frozen PROVIDER_MODEL.
+    # provider_get_current_model() resolves based on LOKI_CURRENT_TIER at runtime.
+    # Falls back to provider_get_tier_param if available, then to GEMINI_DEFAULT_PRO.
+    local model
+    if type provider_get_current_model &>/dev/null; then
+        model=$(provider_get_current_model)
+    else
+        model="${GEMINI_DEFAULT_PRO:-gemini-3-pro-preview}"
+    fi
     local fallback="${PROVIDER_MODEL_FALLBACK:-${GEMINI_DEFAULT_FLASH:-gemini-3-flash-preview}}"
 
     # Create temp file for output to preserve streaming while checking for rate limit
@@ -3057,7 +3080,13 @@ invoke_gemini_capture() {
     local prompt="$1"
     shift
 
-    local model="${PROVIDER_MODEL:-${GEMINI_DEFAULT_PRO:-gemini-3-pro-preview}}"
+    # BUG-PROV-001/006 fix: Use dynamic model resolution instead of frozen PROVIDER_MODEL
+    local model
+    if type provider_get_current_model &>/dev/null; then
+        model=$(provider_get_current_model)
+    else
+        model="${GEMINI_DEFAULT_PRO:-gemini-3-pro-preview}"
+    fi
     local fallback="${PROVIDER_MODEL_FALLBACK:-${GEMINI_DEFAULT_FLASH:-gemini-3-flash-preview}}"
     local output
 
@@ -3249,15 +3278,21 @@ set_phase() {
 
         log_info "Phase changed: $current_phase -> $new_phase"
 
-        # Update orchestrator state
+        # Update orchestrator state (atomic via temp file + mv)
+        # BUG ARCH-001 fix: prevent state corruption if process is killed mid-write
         if [ -f "$orch_file" ]; then
             python3 -c "
-import json, sys
-with open(sys.argv[1], 'r') as f:
+import json, sys, os, tempfile
+orch_file = sys.argv[1]
+new_phase = sys.argv[2]
+with open(orch_file, 'r') as f:
     data = json.load(f)
-data['currentPhase'] = sys.argv[2]
-with open(sys.argv[1], 'w') as f:
+data['currentPhase'] = new_phase
+orch_dir = os.path.dirname(orch_file)
+fd, tmp = tempfile.mkstemp(dir=orch_dir, suffix='.json')
+with os.fdopen(fd, 'w') as f:
     json.dump(data, f, indent=2)
+os.replace(tmp, orch_file)
 " "$orch_file" "$new_phase" 2>/dev/null || true
         fi
     fi
@@ -3627,8 +3662,10 @@ print(json.dumps(data, indent=2))
         fi
     ) 200>"$lockfile"
 
-    # Update current-task.json
-    echo "$task_json" > .loki/queue/current-task.json
+    # BUG-ST-014: Atomic current-task.json update via temp file + mv
+    local ct_tmp=".loki/queue/current-task.json.tmp.$$"
+    echo "$task_json" > "$ct_tmp"
+    mv -f "$ct_tmp" .loki/queue/current-task.json
 }
 
 # Track iteration completion - move task to completed queue
@@ -3811,8 +3848,10 @@ except:
 " 2>/dev/null || true
     fi
 
-    # Clear current-task.json
-    echo "{}" > .loki/queue/current-task.json
+    # BUG-ST-014: Atomic current-task.json clear via temp file + mv
+    local ct_tmp=".loki/queue/current-task.json.tmp.$$"
+    echo "{}" > "$ct_tmp"
+    mv -f "$ct_tmp" .loki/queue/current-task.json
 
     # Write-back completed BMAD stories to source artifacts (v6.29.0)
     if [ "$exit_code" = "0" ]; then
@@ -5526,20 +5565,22 @@ enforce_test_coverage() {
 
     # JavaScript/TypeScript
     if [ -f "${TARGET_DIR:-.}/package.json" ]; then
+        # BUG-EC-014: Wrap test runners with timeout to prevent hanging indefinitely
+        local gate_timeout="${LOKI_GATE_TIMEOUT:-300}"  # 5 minutes default
         if grep -q '"vitest"' "${TARGET_DIR:-.}/package.json" 2>/dev/null; then
             test_runner="vitest"
             local output
-            output=$(cd "${TARGET_DIR:-.}" && npx vitest run --reporter=json 2>&1) || test_passed=false
+            output=$(cd "${TARGET_DIR:-.}" && timeout "$gate_timeout" npx vitest run --reporter=json 2>&1) || test_passed=false
             details="vitest: $(echo "$output" | tail -3 | tr '\n' ' ')"
         elif grep -q '"jest"' "${TARGET_DIR:-.}/package.json" 2>/dev/null; then
             test_runner="jest"
             local output
-            output=$(cd "${TARGET_DIR:-.}" && npx jest --passWithNoTests --forceExit 2>&1) || test_passed=false
+            output=$(cd "${TARGET_DIR:-.}" && timeout "$gate_timeout" npx jest --passWithNoTests --forceExit 2>&1) || test_passed=false
             details="jest: $(echo "$output" | tail -3 | tr '\n' ' ')"
         elif grep -q '"mocha"' "${TARGET_DIR:-.}/package.json" 2>/dev/null; then
             test_runner="mocha"
             local output
-            output=$(cd "${TARGET_DIR:-.}" && npx mocha 2>&1) || test_passed=false
+            output=$(cd "${TARGET_DIR:-.}" && timeout "$gate_timeout" npx mocha 2>&1) || test_passed=false
             details="mocha: $(echo "$output" | tail -3 | tr '\n' ' ')"
         fi
     fi
@@ -6859,16 +6900,32 @@ except Exception:
 PYEOF
 }
 
-# Check provider health: API key exists + CLI installed
+# Check provider health: CLI installed + authentication available
 # Returns: 0 if healthy, 1 if unhealthy
+# BUG-PROV-003 fix: Claude Code supports OAuth sessions in addition to API keys.
+# Checking only for ANTHROPIC_API_KEY incorrectly marks OAuth users as unhealthy,
+# causing unnecessary failover to degraded providers. Now also checks for OAuth
+# session files and `claude auth status` as fallback.
 check_provider_health() {
     local provider="$1"
 
-    # Check CLI is installed
+    # Check CLI is installed and authentication is available
     case "$provider" in
         claude)
             command -v claude &>/dev/null || return 1
-            [ -n "${ANTHROPIC_API_KEY:-}" ] || return 1
+            # Accept API key OR OAuth session (Claude Code supports both)
+            if [ -n "${ANTHROPIC_API_KEY:-}" ]; then
+                return 0
+            fi
+            # Check for OAuth session files (~/.claude/ stores sessions)
+            if [ -d "${HOME}/.claude" ] && [ -f "${HOME}/.claude/.credentials.json" ]; then
+                return 0
+            fi
+            # Last resort: ask the CLI if it has valid auth
+            if claude auth status &>/dev/null 2>&1; then
+                return 0
+            fi
+            return 1
             ;;
         codex)
             command -v codex &>/dev/null || return 1
@@ -6876,7 +6933,15 @@ check_provider_health() {
             ;;
         gemini)
             command -v gemini &>/dev/null || return 1
-            [ -n "${GOOGLE_API_KEY:-${GEMINI_API_KEY:-}}" ] || return 1
+            # BUG-PROV-003: Also accept GEMINI_API_KEY and gcloud ADC
+            if [ -n "${GOOGLE_API_KEY:-}" ] || [ -n "${GEMINI_API_KEY:-}" ]; then
+                return 0
+            fi
+            # Check for gcloud Application Default Credentials
+            if [ -f "${HOME}/.config/gcloud/application_default_credentials.json" ]; then
+                return 0
+            fi
+            return 1
             ;;
         cline)
             command -v cline &>/dev/null || return 1
@@ -6944,7 +7009,13 @@ attempt_provider_failover() {
             update_failover_health "$provider" "healthy"
 
             # Update runtime provider vars
+            # BUG-PROV-008 fix: Update BOTH PROVIDER_NAME and LOKI_PROVIDER.
+            # Without this, subprocesses and the MCP server (which read LOKI_PROVIDER)
+            # continue using the old provider name, causing provider-specific behavior
+            # in child processes to use the wrong config.
             PROVIDER_NAME="$provider"
+            LOKI_PROVIDER="$provider"
+            export LOKI_PROVIDER
 
             emit_event_json "provider_failover" \
                 "from=$current" \
@@ -6996,7 +7067,10 @@ check_primary_recovery() {
         update_failover_state "currentProvider" "$primary"
         update_failover_health "$primary" "healthy"
 
+        # BUG-PROV-008 fix: Update BOTH PROVIDER_NAME and LOKI_PROVIDER on recovery
         PROVIDER_NAME="$primary"
+        LOKI_PROVIDER="$primary"
+        export LOKI_PROVIDER
 
         emit_event_json "provider_recovery" \
             "from=$current" \
@@ -7935,6 +8009,9 @@ save_state() {
     local status="$2"
     local exit_code="$3"
 
+    # BUG-ST-013: Ensure .loki directory exists (defensive -- may be called from signal handler)
+    mkdir -p .loki 2>/dev/null || true
+
     # BUG-XC-004: Atomic write via temp file + mv
     local state_tmp=".loki/autonomy-state.json.tmp.$$"
     cat > "$state_tmp" << EOF
@@ -7954,8 +8031,44 @@ EOF
 }
 
 load_state() {
+    # BUG-EP-015: Clean up orphaned temp files from kill -9 crashes
+    # These are left behind when the process is killed during atomic writes
+    find .loki/ -maxdepth 1 -name "*.tmp.*" -mmin +5 -delete 2>/dev/null || true
+    find .loki/state/ -name "*.tmp.*" -mmin +5 -delete 2>/dev/null || true
+
     if [ -f ".loki/autonomy-state.json" ]; then
         if command -v python3 &> /dev/null; then
+            # BUG-ST-006: Validate checkpoint integrity before loading state
+            local state_valid
+            state_valid=$(python3 -c "
+import json, sys
+try:
+    with open('.loki/autonomy-state.json') as f:
+        d = json.load(f)
+    # Validate required fields exist and have sane types
+    rc = d.get('retryCount', 0)
+    ic = d.get('iterationCount', 0)
+    status = d.get('status', 'unknown')
+    if not isinstance(rc, (int, float)) or not isinstance(ic, (int, float)):
+        print('invalid')
+        sys.exit(0)
+    if rc < 0 or ic < 0:
+        print('invalid')
+        sys.exit(0)
+    print('valid')
+except (json.JSONDecodeError, KeyError, TypeError, OSError):
+    print('invalid')
+" 2>/dev/null || echo "invalid")
+
+            if [ "$state_valid" != "valid" ]; then
+                log_warn "State file corrupted or invalid - starting fresh"
+                RETRY_COUNT=0
+                ITERATION_COUNT=0
+                # Back up corrupted state file for diagnosis
+                mv ".loki/autonomy-state.json" ".loki/autonomy-state.json.corrupt.$(date +%s)" 2>/dev/null || true
+                return
+            fi
+
             # Load retry count, iteration count, and status from previous session
             local prev_status
             prev_status=$(python3 -c "import json; print(json.load(open('.loki/autonomy-state.json')).get('status', 'unknown'))" 2>/dev/null || echo "unknown")
@@ -9280,16 +9393,9 @@ run_autonomous() {
     fi
 
     while [ $retry -lt $MAX_RETRIES ]; do
-        # Increment iteration count
-        ((ITERATION_COUNT++))
-
-        # Check max iterations
-        if check_max_iterations; then
-            save_state $retry "max_iterations_reached" 0
-            return 0
-        fi
-
-        # Check for human intervention (PAUSE, HUMAN_INPUT.md, STOP)
+        # Check for human intervention BEFORE incrementing iteration count
+        # BUG-ST-010: Moved pause/stop checks before ITERATION_COUNT increment
+        # to prevent spurious count increases when resuming from pause
         check_human_intervention
         local intervention_result=$?
         case $intervention_result in
@@ -9304,6 +9410,15 @@ run_autonomous() {
             continue  # Will hit PAUSE check on next iteration
         fi
 
+        # Increment iteration count (after pause/stop checks to avoid spurious increments)
+        ((ITERATION_COUNT++))
+
+        # Check max iterations
+        if check_max_iterations; then
+            save_state $retry "max_iterations_reached" 0
+            return 0
+        fi
+
         # Watchdog: periodic process health check (opt-in via LOKI_WATCHDOG=true)
         if [[ "$WATCHDOG_ENABLED" == "true" ]]; then
             local now_epoch
@@ -9375,6 +9490,11 @@ run_autonomous() {
 
         # Dynamic tier selection based on RARV cycle phase
         CURRENT_TIER=$(get_rarv_tier "$ITERATION_COUNT")
+        # NEW BUG FIX: Export LOKI_CURRENT_TIER so provider helper functions
+        # (e.g., gemini.sh:provider_get_current_model) can resolve the correct model.
+        # Without this, LOKI_CURRENT_TIER is always empty and defaults to "planning".
+        LOKI_CURRENT_TIER="$CURRENT_TIER"
+        export LOKI_CURRENT_TIER
         local rarv_phase=$(get_rarv_phase_name "$ITERATION_COUNT")
         local tier_param=$(get_provider_tier_param "$CURRENT_TIER")
         echo "=== RARV Phase: $rarv_phase, Tier: $CURRENT_TIER ($tier_param) ===" | tee -a "$log_file" "$agent_log"
@@ -9635,25 +9755,46 @@ if __name__ == "__main__":
             gemini)
                 # Gemini: Degraded mode - no stream-json, no agent tracking
                 # Uses invoke_gemini helper for rate limit fallback to flash model
-                local model="${PROVIDER_MODEL:-${GEMINI_DEFAULT_PRO:-gemini-3-pro-preview}}"
+                # BUG-PROV-001 fix: Use tier_param (resolved model) instead of frozen PROVIDER_MODEL
+                # tier_param is computed above via get_provider_tier_param() -> resolve_model_for_tier()
+                # which returns the correct model name for the current RARV tier
+                local model="$tier_param"
                 local fallback="${PROVIDER_MODEL_FALLBACK:-${GEMINI_DEFAULT_FLASH:-gemini-3-flash-preview}}"
-                echo "[loki] Gemini model: $model (fallback: $fallback), tier: $tier_param" >> "$log_file"
-                echo "[loki] Gemini model: $model (fallback: $fallback), tier: $tier_param" >> "$agent_log"
+                echo "[loki] Gemini model: $model (fallback: $fallback), tier: $CURRENT_TIER" >> "$log_file"
+                echo "[loki] Gemini model: $model (fallback: $fallback), tier: $CURRENT_TIER" >> "$agent_log"
+
+                # BUG-PROV-003: Resolve API key (supports GEMINI_API_KEY alias and ADC)
+                if type _gemini_resolve_api_key &>/dev/null; then
+                    _gemini_resolve_api_key || true
+                fi
 
-                # Try primary model, fallback on rate limit
-                local tmp_output
+                # Try primary model, fallback on rate limit or auth error
+                local tmp_output tmp_stderr
                 tmp_output=$(mktemp)
+                tmp_stderr=$(mktemp)
                 # BUG-RUN-011/RUN-013: Use PIPESTATUS[0] for primary invocation too
-                gemini --approval-mode=yolo --model "$model" "$prompt" < /dev/null 2>&1 | tee "$tmp_output" | tee -a "$log_file" "$agent_log" "$iter_output"
+                gemini --approval-mode=yolo --model "$model" "$prompt" < /dev/null 2>"$tmp_stderr" | tee "$tmp_output" | tee -a "$log_file" "$agent_log" "$iter_output"
                 exit_code=${PIPESTATUS[0]}
 
-                if [[ $exit_code -ne 0 ]] && grep -qiE "(rate.?limit|429|quota|resource.?exhausted)" "$tmp_output"; then
+                # BUG-PROV-003: Handle auth errors with API key rotation
+                if [[ $exit_code -ne 0 ]] && grep -qiE "(401|403|unauthorized|forbidden|invalid.?api.?key|permission.?denied)" "$tmp_stderr" 2>/dev/null; then
+                    if type _gemini_rotate_api_key &>/dev/null && _gemini_rotate_api_key; then
+                        log_warn "Auth error on Gemini, rotated to next API key"
+                        rm -f "$tmp_output" "$tmp_stderr"
+                        tmp_output=$(mktemp)
+                        tmp_stderr=$(mktemp)
+                        gemini --approval-mode=yolo --model "$model" "$prompt" < /dev/null 2>"$tmp_stderr" | tee "$tmp_output" | tee -a "$log_file" "$agent_log" "$iter_output"
+                        exit_code=${PIPESTATUS[0]}
+                    fi
+                fi
+
+                if [[ $exit_code -ne 0 ]] && grep -qiE "(rate.?limit|429|quota|resource.?exhausted)" "$tmp_stderr" "$tmp_output" 2>/dev/null; then
                     log_warn "Rate limit hit on $model, falling back to $fallback"
                     echo "[loki] Fallback to $fallback due to rate limit" >> "$log_file"
                     gemini --approval-mode=yolo --model "$fallback" "$prompt" < /dev/null 2>&1 | tee -a "$log_file" "$agent_log" "$iter_output"
                     exit_code=${PIPESTATUS[0]}
                 fi
-                rm -f "$tmp_output"
+                rm -f "$tmp_output" "$tmp_stderr"
                 ;;
 
             cline)
@@ -9688,6 +9829,13 @@ if __name__ == "__main__":
         local duration=$((end_time - start_time))
 
         log_info "${PROVIDER_DISPLAY_NAME:-Claude} exited with code $exit_code after ${duration}s"
+
+        # BUG-EC-013: Detect empty provider output (0 bytes = no work done)
+        if [ -f "$iter_output" ] && [ ! -s "$iter_output" ] && [ $exit_code -eq 0 ]; then
+            log_warn "Provider returned empty output (0 bytes) despite exit code 0 -- treating as error"
+            exit_code=1
+        fi
+
         save_state $retry "exited" $exit_code
 
         # Auto-track iteration completion (for dashboard task queue)
@@ -9775,6 +9923,16 @@ if __name__ == "__main__":
                     log_warn "Static analysis FAILED ($sa_count consecutive) - findings injected into next iteration"
                 fi
             fi
+            # BUG-ST-002: Check pause signal between quality gates
+            if [ -f "${TARGET_DIR:-.}/.loki/PAUSE" ] || [ -f "${TARGET_DIR:-.}/.loki/STOP" ]; then
+                log_warn "Pause/stop signal detected between quality gates - deferring remaining gates"
+                # Store partial gate failures before breaking out
+                if [ -n "$gate_failures" ]; then
+                    echo "$gate_failures" > "${TARGET_DIR:-.}/.loki/quality/gate-failures.txt"
+                fi
+                # Let the main loop handle the pause/stop on next iteration
+                continue
+            fi
             # Test coverage gate
             if [ "${PHASE_UNIT_TESTS:-true}" = "true" ]; then
                 log_info "Quality gate: test coverage..."
@@ -9787,6 +9945,14 @@ if __name__ == "__main__":
                     log_warn "Test coverage gate FAILED ($tc_count consecutive) - must pass next iteration"
                 fi
             fi
+            # BUG-ST-002: Check pause signal between quality gates (after test coverage)
+            if [ -f "${TARGET_DIR:-.}/.loki/PAUSE" ] || [ -f "${TARGET_DIR:-.}/.loki/STOP" ]; then
+                log_warn "Pause/stop signal detected between quality gates - deferring remaining gates"
+                if [ -n "$gate_failures" ]; then
+                    echo "$gate_failures" > "${TARGET_DIR:-.}/.loki/quality/gate-failures.txt"
+                fi
+                continue
+            fi
             # Code review gate (upgraded from advisory, with escalation)
             if [ "$PHASE_CODE_REVIEW" = "true" ] && [ "$ITERATION_COUNT" -gt 0 ]; then
                 log_info "Quality gate: code review..."
@@ -9850,6 +10016,8 @@ if __name__ == "__main__":
                 log_info "Perpetual mode: Ignoring exit, continuing immediately..."
                 # BUG-RUN-010: Reset retry counter on success (only count failures)
                 retry=0
+                # BUG-NEW-003/E2E-005: Clean up per-iteration output before continuing
+                rm -f "$iter_output" 2>/dev/null
                 continue  # Immediately start next iteration, no wait
             fi
 
@@ -9896,6 +10064,8 @@ if __name__ == "__main__":
             log_step "Starting next iteration..."
             # BUG-RUN-010: Reset retry counter on success (only count failures)
             retry=0
+            # BUG-NEW-003/E2E-005: Clean up per-iteration output before continuing
+            rm -f "$iter_output" 2>/dev/null
             continue  # Immediately start next iteration, no exponential backoff
         fi
 
@@ -10109,9 +10279,18 @@ check_human_intervention() {
 
 # Handle pause state - wait for resume
 handle_pause() {
+    # BUG-ST-007: Guard against concurrent pause handler execution
+    if [ "${_PAUSE_IN_PROGRESS:-0}" -eq 1 ]; then
+        return 0
+    fi
+    _PAUSE_IN_PROGRESS=1
+
     PAUSED=true
     local loki_dir="${TARGET_DIR:-.}/.loki"
 
+    # Save state before pausing so it persists across potential crashes
+    save_state ${RETRY_COUNT:-0} "paused" 0
+
     log_header "Execution Paused"
     echo ""
     log_info "To resume: Remove .loki/PAUSE or press Enter"
@@ -10141,6 +10320,7 @@ EOF
         if [ -f "$loki_dir/STOP" ]; then
             rm -f "$loki_dir/STOP" "$loki_dir/PAUSED.md"
             PAUSED=false
+            _PAUSE_IN_PROGRESS=0
             return 1
         fi
 
@@ -10163,6 +10343,7 @@ EOF
     rm -f "$loki_dir/PAUSED.md"
     log_info "Resuming execution..."
     PAUSED=false
+    _PAUSE_IN_PROGRESS=0
     return 0
 }
 

package/dashboard/__init__.py

@@ -7,7 +7,7 @@ Modules:
     control: Session control API (start/stop/pause/resume)
 """
 
-__version__ = "6.71.1"
+__version__ = "6.72.0"
 
 # Expose the control app for easy import
 try: