@zachjxyz/moxie 0.2.4 → 0.3.0

package/lib/agents.sh

@@ -1,12 +1,146 @@
 #!/usr/bin/env bash
-# moxie/lib/agents.sh — Agent dispatch and rotation
+# moxie/lib/agents.sh — Agent registry, dispatch, rotation, and health checks
 # Compatible with Bash 3.2 (macOS default) — no associative arrays.
 
+# ---- Known agent registry ----
+# To add a new agent: append one entry to each array at the same index.
+# The command template is what gets written to config.toml. The prompt is
+# appended as the final argument by dispatch_agent().
+
+KNOWN_AGENT_NAMES=(claude codex qwen aider goose amp cline roo)
+KNOWN_AGENT_LABELS=("Claude Code" "Codex" "Qwen Code" "Aider" "Goose" "Amp" "Cline" "Roo Code")
+KNOWN_AGENT_BINARIES=(claude codex qwen aider goose amp cline roo)
+KNOWN_AGENT_CMDS=(
+    'claude --dangerously-skip-permissions --effort max --output-format json -p'
+    'codex exec -c model_reasoning_effort="xhigh" --dangerously-bypass-approvals-and-sandbox -c model_reasoning_summary="detailed" -c model_supports_reasoning_summaries=true'
+    'qwen --yolo -o json'
+    'aider --yes-always -m'
+    'goose run --output-format json -t'
+    'amp --dangerously-allow-all --stream-json -x'
+    'cline -y --json'
+    'roo --output-format json'
+)
+
+# ---- Known gateway models (Vercel AI Gateway) ----
+# To add a new gateway model: append one entry to each array at the same index.
+
+KNOWN_GATEWAY_NAMES=(claude-gw gpt-gw qwen-gw glm-gw kimi-gw)
+KNOWN_GATEWAY_MODELS=(
+    "anthropic/claude-opus-4-6"
+    "openai/gpt-5.4"
+    "qwen/qwen3.6-plus"
+    "zai/glm-5.1"
+    "moonshotai/kimi-k2.5"
+)
+KNOWN_GATEWAY_LABELS=(
+    "Claude Opus 4.6"
+    "GPT-5.4"
+    "Qwen 3.6 Plus"
+    "GLM 5.1"
+    "Kimi 2.5"
+)
+
+# ---- Gateway agent helpers ----
+
+# Returns 0 if the agent is a gateway agent (command starts with "gateway:")
+_is_gateway_agent() {
+    local name="$1"
+    local cmd
+    cmd=$(_agent_cmd "$name")
+    case "$cmd" in
+        gateway:*) return 0 ;;
+        *) return 1 ;;
+    esac
+}
+
+# Extracts the model string from a gateway agent's stored command
+_agent_model() {
+    local name="$1"
+    local cmd
+    cmd=$(_agent_cmd "$name")
+    echo "${cmd#gateway:}"
+}
+
+# ---- Detect available agents on PATH ----
+
+detect_available_agents() {
+    AVAILABLE_AGENT_INDICES=()
+    AVAILABLE_AGENT_COUNT=0
+
+    for i in "${!KNOWN_AGENT_BINARIES[@]}"; do
+        if command -v "${KNOWN_AGENT_BINARIES[$i]}" &>/dev/null; then
+            AVAILABLE_AGENT_INDICES+=("$i")
+        fi
+    done
+    AVAILABLE_AGENT_COUNT=${#AVAILABLE_AGENT_INDICES[@]}
+}
+
+# ---- Check minimum agent count (from config) ----
+
+check_minimum_agents() {
+    if [ "${AGENT_COUNT:-0}" -eq 0 ]; then
+        load_agents
+    fi
+
+    local healthy=0
+    for i in "${!AGENT_NAMES[@]}"; do
+        local name="${AGENT_NAMES[$i]}"
+        if _is_gateway_agent "$name"; then
+            # Gateway agent: needs node + stored key
+            if command -v node &>/dev/null && gateway_has_key "vercel-ai-gateway"; then
+                healthy=$((healthy + 1))
+            fi
+        else
+            local cmd
+            cmd=$(_agent_cmd "$name")
+            local binary
+            binary=$(echo "$cmd" | awk '{print $1}')
+            if command -v "$binary" &>/dev/null; then
+                healthy=$((healthy + 1))
+            fi
+        fi
+    done
+
+    if [ "$healthy" -lt 2 ]; then
+        echo "" >&2
+        echo "ERROR: moxie requires at least 2 healthy agents for cross-model verification." >&2
+        echo "Single-agent mode is not supported." >&2
+        echo "" >&2
+        echo "Configured agents:" >&2
+        for i in "${!AGENT_NAMES[@]}"; do
+            local name="${AGENT_NAMES[$i]}"
+            if _is_gateway_agent "$name"; then
+                local model
+                model=$(_agent_model "$name")
+                if command -v node &>/dev/null && gateway_has_key "vercel-ai-gateway"; then
+                    echo "  [OK] $name — $model (AI Gateway)" >&2
+                else
+                    echo "  [!!] $name — $model (gateway key missing or node not found)" >&2
+                fi
+            else
+                local cmd
+                cmd=$(_agent_cmd "$name")
+                local binary
+                binary=$(echo "$cmd" | awk '{print $1}')
+                if command -v "$binary" &>/dev/null; then
+                    echo "  [OK] $name" >&2
+                else
+                    echo "  [!!] $name — '$binary' not found on PATH" >&2
+                fi
+            fi
+        done
+        echo "" >&2
+        echo "Run 'moxie doctor' for details, or 'moxie init' to reconfigure." >&2
+        return 1
+    fi
+    return 0
+}
+
 # Global arrays populated by load_agents():
 #   AGENT_NAMES=("codex" "claude" "qwen")
-#   AGENT_CMDS_0="codex exec ..."  (indexed by position in AGENT_NAMES)
-#   AGENT_CMDS_1="claude ..."
-#   AGENT_CMDS_2="qwen ..."
+#   AGENT_CMD_0="codex exec ..."  (indexed by position in AGENT_NAMES)
+#   AGENT_CMD_1="claude ..."
+#   AGENT_CMD_2="qwen ..."
 
 # ---- Load agents from config ----
 
@@ -20,30 +154,40 @@ load_agents() {
     local current_agent=""
     local current_cmd=""
     local current_order="99"
+    local current_type=""
+    local current_model=""
+
+    _commit_agent() {
+        if [ -z "$current_agent" ]; then return; fi
+        # CLI agent: has command. Gateway agent: type=gateway with model.
+        if [ -n "$current_cmd" ]; then
+            AGENT_NAMES+=("$current_agent")
+            agent_cmds+=("$current_cmd")
+            agent_orders+=("$current_order")
+        elif [ "$current_type" = "gateway" ] && [ -n "$current_model" ]; then
+            AGENT_NAMES+=("$current_agent")
+            agent_cmds+=("gateway:$current_model")
+            agent_orders+=("$current_order")
+        fi
+    }
 
     while IFS= read -r line; do
         [[ "$line" =~ ^[[:space:]]*# ]] && continue
         [[ -z "${line// /}" ]] && continue
 
         if [[ "$line" =~ ^\[agents\.([a-zA-Z0-9_-]+)\] ]]; then
-            if [ -n "$current_agent" ] && [ -n "$current_cmd" ]; then
-                AGENT_NAMES+=("$current_agent")
-                agent_cmds+=("$current_cmd")
-                agent_orders+=("$current_order")
-            fi
+            _commit_agent
             current_agent="${BASH_REMATCH[1]}"
             current_cmd=""
             current_order="99"
+            current_type=""
+            current_model=""
             continue
         fi
 
         # New non-agent section
         if [[ "$line" =~ ^\[ ]] && ! [[ "$line" =~ ^\[agents\. ]]; then
-            if [ -n "$current_agent" ] && [ -n "$current_cmd" ]; then
-                AGENT_NAMES+=("$current_agent")
-                agent_cmds+=("$current_cmd")
-                agent_orders+=("$current_order")
-            fi
+            _commit_agent
             current_agent=""
             continue
         fi
@@ -59,15 +203,21 @@ load_agents() {
             if [[ "$line" =~ ^[[:space:]]*order[[:space:]]*=[[:space:]]*([0-9]+) ]]; then
                 current_order="${BASH_REMATCH[1]}"
             fi
+            if [[ "$line" =~ ^[[:space:]]*type[[:space:]]*=[[:space:]]*(.*) ]]; then
+                current_type="${BASH_REMATCH[1]}"
+                current_type="${current_type#\"}"
+                current_type="${current_type%\"}"
+            fi
+            if [[ "$line" =~ ^[[:space:]]*model[[:space:]]*=[[:space:]]*(.*) ]]; then
+                current_model="${BASH_REMATCH[1]}"
+                current_model="${current_model#\"}"
+                current_model="${current_model%\"}"
+            fi
         fi
     done < "$MOXIE_CONFIG"
 
     # Save last agent
-    if [ -n "$current_agent" ] && [ -n "$current_cmd" ]; then
-        AGENT_NAMES+=("$current_agent")
-        agent_cmds+=("$current_cmd")
-        agent_orders+=("$current_order")
-    fi
+    _commit_agent
 
     # Sort by order using a temp file
     local sorted
@@ -110,6 +260,7 @@ dispatch_agent() {
     local agent="$1"
     local prompt_file="$2"
     local timeout_secs="$3"
+    local phase="${4:-unknown}"
     local cmd
     cmd=$(_agent_cmd "$agent")
 
@@ -118,7 +269,34 @@ dispatch_agent() {
         return 1
     fi
 
-    # Isolate: strip plugin root vars so hooks from other CLIs can't fire
+    # Gateway agent: dispatch via Node.js script
+    if _is_gateway_agent "$agent"; then
+        local model endpoint api_key
+        model=$(_agent_model "$agent")
+        endpoint=$(toml_get "$MOXIE_CONFIG" "gateway.endpoint" "https://ai-gateway.vercel.sh")
+        api_key=$(gateway_get_key "vercel-ai-gateway") || {
+            echo "ERROR: Gateway API key not found. Run 'moxie init' to configure." >&2
+            return 1
+        }
+
+        env -u CLAUDE_PLUGIN_ROOT -u CODEX_PLUGIN_ROOT -u CURSOR_PLUGIN_ROOT \
+            GATEWAY_API_KEY="$api_key" \
+            GATEWAY_ENDPOINT="$endpoint" \
+            GATEWAY_MODEL="$model" \
+            GATEWAY_PHASE="$phase" \
+            GATEWAY_AGENT="$agent" \
+            timeout "$timeout_secs" \
+            node "$MOXIE_ROOT/lib/gateway-agent.mjs" "$(cat "$prompt_file")" || {
+            local rc=$?
+            if [ $rc -eq 124 ]; then
+                echo "[TIMEOUT] $agent exceeded ${timeout_secs}s — skipping"
+            fi
+            return $rc
+        }
+        return 0
+    fi
+
+    # CLI agent: dispatch via shell command
     env -u CLAUDE_PLUGIN_ROOT \
         -u CODEX_PLUGIN_ROOT \
         -u CURSOR_PLUGIN_ROOT \
@@ -145,6 +323,109 @@ next_agent() {
     echo "${AGENT_NAMES[0]}"
 }
 
+# ---- Agent health tracking ----
+# Tracks consecutive failures per agent. An agent is degraded after
+# AGENT_DEGRADE_THRESHOLD consecutive non-productive turns (timeout,
+# empty output, or no useful work done).
+
+AGENT_DEGRADE_THRESHOLD=2
+
+# Parallel arrays indexed by AGENT_NAMES position.
+# Initialized by _init_agent_health() after load_agents.
+AGENT_FAIL_STREAK=()
+AGENT_DEGRADED=()
+
+_init_agent_health() {
+    AGENT_FAIL_STREAK=()
+    AGENT_DEGRADED=()
+    for (( i = 0; i < AGENT_COUNT; i++ )); do
+        AGENT_FAIL_STREAK+=(0)
+        AGENT_DEGRADED+=(0)
+    done
+}
+
+# Returns 0 if the agent is degraded, 1 otherwise.
+is_agent_degraded() {
+    local name="$1"
+    for i in "${!AGENT_NAMES[@]}"; do
+        if [ "${AGENT_NAMES[$i]}" = "$name" ]; then
+            [ "${AGENT_DEGRADED[$i]}" = "1" ]
+            return
+        fi
+    done
+    return 1
+}
+
+# Record a productive or non-productive turn for an agent.
+# Usage: _record_turn_health <agent_name> <logfile>
+_record_turn_health() {
+    local name="$1"
+    local logfile="$2"
+
+    local idx=-1
+    for i in "${!AGENT_NAMES[@]}"; do
+        [ "${AGENT_NAMES[$i]}" = "$name" ] && idx=$i && break
+    done
+    [ "$idx" = "-1" ] && return
+
+    # Already degraded — don't re-evaluate
+    [ "${AGENT_DEGRADED[$idx]}" = "1" ] && return
+
+    # A productive turn has a log file > 1KB with actual agent output
+    # (not just the prompt echo or timeout message)
+    local log_size=0
+    if [ -f "$logfile" ]; then
+        log_size=$(wc -c < "$logfile" | tr -d ' ')
+    fi
+
+    # Check for known failure signatures in the log
+    local is_failure=0
+    if [ "$log_size" -lt 1024 ]; then
+        is_failure=1
+    elif grep -qE '^\[TIMEOUT\]' "$logfile" 2>/dev/null; then
+        is_failure=1
+    fi
+
+    if [ "$is_failure" = "1" ]; then
+        AGENT_FAIL_STREAK[$idx]=$(( ${AGENT_FAIL_STREAK[$idx]} + 1 ))
+        local streak=${AGENT_FAIL_STREAK[$idx]}
+
+        if [ "$streak" -ge "$AGENT_DEGRADE_THRESHOLD" ]; then
+            AGENT_DEGRADED[$idx]=1
+            echo "  [DEGRADED] $name failed $streak consecutive turns — removing from rotation"
+            echo "             (possible quota exhaustion, auth failure, or CLI crash)"
+
+            # Count remaining healthy agents
+            local healthy=0
+            for j in "${!AGENT_DEGRADED[@]}"; do
+                [ "${AGENT_DEGRADED[$j]}" = "0" ] && healthy=$((healthy + 1))
+            done
+
+            if [ "$healthy" -lt 2 ]; then
+                echo ""
+                echo "  FATAL: Only $healthy healthy agent(s) remaining."
+                echo "  moxie requires at least 2 agents for cross-model verification."
+                echo "  Pipeline cannot continue. Stopping."
+                return 2
+            fi
+        else
+            echo "  [WARN] $name produced no useful output (turn $streak/$AGENT_DEGRADE_THRESHOLD before degraded)"
+        fi
+    else
+        # Productive turn — reset streak
+        AGENT_FAIL_STREAK[$idx]=0
+    fi
+}
+
+# Count healthy (non-degraded) agents.
+count_healthy_agents() {
+    local healthy=0
+    for i in "${!AGENT_DEGRADED[@]}"; do
+        [ "${AGENT_DEGRADED[$i]}" = "0" ] && healthy=$((healthy + 1))
+    done
+    echo "$healthy"
+}
+
 # ---- Logging ----
 
 dispatch_logged() {
@@ -168,12 +449,15 @@ dispatch_logged() {
         return 0
     fi
 
-    dispatch_agent "$agent" "$prompt_file" "$timeout_secs" 2>&1 | tee "$logfile" || true
+    dispatch_agent "$agent" "$prompt_file" "$timeout_secs" "$phase" 2>&1 | tee "$logfile" || true
 
     local tokens
     tokens=$(extract_tokens "$logfile" "$agent")
     record_tokens "$csv_file" "$turn" "$agent" "$ts" "$tokens" "$phase"
     echo "  Tokens: $tokens"
+
+    # Track agent health based on output
+    _record_turn_health "$agent" "$logfile" || return $?
 }
 
 # ---- List agents ----
@@ -210,13 +494,18 @@ cmd_doctor() {
     # ---- Dependencies ----
     echo "Dependencies:"
     _check_dep "python3" "Required for ledger parsing and token tracking" || all_ok=0
-    _check_dep "caffeinate" "Keeps machine awake during runs (macOS)" || true  # non-fatal
+    # Sleep inhibitor: platform-specific, non-fatal
+    if [ "$MOXIE_PLATFORM" = "darwin" ]; then
+        _check_dep "caffeinate" "Keeps machine awake during runs (macOS)" || true
+    elif [ "$MOXIE_PLATFORM" = "linux" ]; then
+        _check_dep "systemd-inhibit" "Keeps machine awake during runs (Linux)" || true
+    fi
     if command -v timeout &>/dev/null; then
         echo "  [OK] timeout"
     elif command -v gtimeout &>/dev/null; then
         echo "  [OK] coreutils (gtimeout aliased as timeout)"
     else
-        echo "  [!!] coreutils — not found. Install: brew install coreutils"
+        echo "  [!!] coreutils — not found. Install: $(coreutils_install_hint)"
         echo "        Provides timeout protection for agent turns (gtimeout)."
         all_ok=0
     fi
@@ -225,37 +514,78 @@ cmd_doctor() {
     # ---- Agents ----
     if [ "$has_project" = "1" ]; then
         echo "Agents (from .moxie/config.toml):"
+        local healthy_count=0
         for i in "${!AGENT_NAMES[@]}"; do
             local name="${AGENT_NAMES[$i]}"
-            local cmd
-            cmd=$(_agent_cmd "$name")
 
-            # Extract the base binary from the command
-            local binary
-            binary=$(echo "$cmd" | awk '{print $1}')
-
-            if ! command -v "$binary" &>/dev/null; then
-                echo "  [!!] $name — '$binary' not found on PATH"
-                all_ok=0
-                continue
-            fi
-
-            # Try --version to verify it's functional
-            local version
-            version=$("$binary" --version 2>&1 | head -1)
-            if [ $? -eq 0 ] && [ -n "$version" ]; then
-                echo "  [OK] $name — $version"
+            if _is_gateway_agent "$name"; then
+                # Gateway agent
+                local model
+                model=$(_agent_model "$name")
+                local gw_ok=1
+
+                if ! command -v node &>/dev/null; then
+                    echo "  [!!] $name — $model (node not found on PATH)"
+                    all_ok=0
+                    gw_ok=0
+                fi
+
+                if [ "$gw_ok" = "1" ] && ! gateway_has_key "vercel-ai-gateway"; then
+                    echo "  [!!] $name — $model (AI Gateway key not configured)"
+                    echo "        Run 'moxie init' to set up your gateway key."
+                    all_ok=0
+                    gw_ok=0
+                fi
+
+                if [ "$gw_ok" = "1" ]; then
+                    echo "  [OK] $name — $model (AI Gateway, key stored)"
+                    healthy_count=$((healthy_count + 1))
+                fi
             else
-                echo "  [??] $name — '$binary' found but --version failed"
-                echo "        May not be authenticated. Try running: $binary --version"
+                # CLI agent
+                local cmd
+                cmd=$(_agent_cmd "$name")
+                local binary
+                binary=$(echo "$cmd" | awk '{print $1}')
+
+                if ! command -v "$binary" &>/dev/null; then
+                    echo "  [!!] $name — '$binary' not found on PATH"
+                    all_ok=0
+                    continue
+                fi
+
+                healthy_count=$((healthy_count + 1))
+
+                local version
+                version=$("$binary" --version 2>&1 | head -1)
+                if [ $? -eq 0 ] && [ -n "$version" ]; then
+                    echo "  [OK] $name — $version"
+                else
+                    echo "  [??] $name — '$binary' found but --version failed"
+                    echo "        May not be authenticated. Try running: $binary --version"
+                fi
             fi
         done
+
+        if [ "$healthy_count" -lt 2 ]; then
+            echo ""
+            echo "  [FAIL] Only $healthy_count healthy agent(s). moxie requires at least 2"
+            echo "         for cross-model verification. Single-agent mode is not supported."
+            all_ok=0
+        fi
     else
         echo "Agents:"
-        echo "  (no .moxie/ project — checking common agents)"
-        _check_agent_binary "codex" || all_ok=0
-        _check_agent_binary "claude" || all_ok=0
-        _check_agent_binary "qwen" || all_ok=0
+        echo "  (no .moxie/ project — scanning for known agents)"
+        detect_available_agents
+        for i in "${!KNOWN_AGENT_NAMES[@]}"; do
+            _check_agent_binary "${KNOWN_AGENT_BINARIES[$i]}" || true
+        done
+        if [ "$AVAILABLE_AGENT_COUNT" -lt 2 ]; then
+            echo ""
+            echo "  [FAIL] Only $AVAILABLE_AGENT_COUNT agent(s) found. moxie requires at least 2"
+            echo "         for cross-model verification. Single-agent mode is not supported."
+            all_ok=0
+        fi
     fi
     echo ""
 
@@ -295,8 +625,10 @@ cmd_doctor() {
     echo ""
     if [ "$all_ok" = "1" ]; then
         echo "All checks passed. Ready to run."
+        return 0
     else
         echo "Some checks failed. Fix the issues above before running."
+        return 1
     fi
 }