a2zusage 1.0.8 → 1.0.9

package/README.md

@@ -78,7 +78,7 @@ a2zusage supports **14+ AI coding tools** out of the box:
 | **Cline / Roo Code** | VS Code extension storage | ✅ Exact token counts (when stored by the extension) |
 | **OpenCode** | Local JSON files | ✅ Exact token counts (when present in session/message usage fields) |
 | **OpenAI Codex** | OpenAI Usage API | ✅ Exact token counts (requires API key + org access) |
-| **Gemini CLI** | Local telemetry (`~/.gemini/`) | ⚠️ Partial: JSON/JSONL telemetry supported when present; protobuf `.pb` logs are detected but not yet parsed |
+|| **Gemini CLI** | Local telemetry (`~/.gemini/`) | ✅ Exact token counts with wrapper (see [Gemini CLI Setup](#gemini-cli-setup)) |
 | **Amazon Q Developer** | Local logs | ⚠️ Best-effort: logs may not contain token totals |
 | **Tabnine** | Local logs | ⚠️ Partial: uses explicit token fields when present; no invented prompt/context tokens |
 | **Gemini Code Assist** | Google Cloud | ⚠️ Not implemented in this repo yet |
@@ -141,6 +141,26 @@ a2zusage -v          # Verbose mode with data sources
 ]
 ```
 
+## Tool-Specific Setup
+
+### Gemini CLI Setup
+
+Gemini CLI stores conversation data in encrypted protobuf files. To get accurate token tracking, use our wrapper script that captures the `usageMetadata` from the `--output-format stream-json` output:
+
+```bash
+# 1. Copy the wrapper script
+curl -o ~/.local/bin/gemini-wrapper https://raw.githubusercontent.com/aezizhu/a2zaiusage/main/scripts/gemini-wrapper.sh
+chmod +x ~/.local/bin/gemini-wrapper
+
+# 2. Create an alias (add to your .bashrc or .zshrc)
+alias gemini="~/.local/bin/gemini-wrapper"
+
+# 3. Use gemini as normal - token data is now tracked!
+gemini "Your prompt here"
+```
+
+The wrapper creates `~/.gemini/a2zusage-telemetry.jsonl` with exact token counts for each request.
+
 ## Why This Tool?
 
 ### 🎯 Built for Hiring AI-Native Developers

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "a2zusage",
-  "version": "1.0.8",
+  "version": "1.0.9",
   "description": "Query usage statistics from all your AI coding tools in one command - Claude Code, Cursor, GitHub Copilot, Windsurf, Warp AI, and 14+ tools",
   "keywords": [
     "ai-usage",

package/scripts/gemini-wrapper.sh

@@ -0,0 +1,87 @@
+#!/bin/bash
+# Gemini CLI Wrapper for a2zusage
+# This script wraps the Gemini CLI to capture usage statistics
+#
+# Installation:
+#   1. Make this script executable: chmod +x gemini-wrapper.sh
+#   2. Add alias to your shell: alias gemini='/path/to/gemini-wrapper.sh'
+#   Or symlink: ln -s /path/to/gemini-wrapper.sh /usr/local/bin/gemini-tracked
+
+USAGE_FILE="${HOME}/.gemini/a2zusage-telemetry.jsonl"
+REAL_GEMINI=$(which -a gemini | grep -v "gemini-wrapper" | head -1)
+
+# If no real gemini found, try common paths
+if [ -z "$REAL_GEMINI" ]; then
+    for path in /opt/homebrew/bin/gemini /usr/local/bin/gemini ~/.local/bin/gemini; do
+        if [ -x "$path" ]; then
+            REAL_GEMINI="$path"
+            break
+        fi
+    done
+fi
+
+if [ -z "$REAL_GEMINI" ]; then
+    echo "Error: Could not find gemini CLI" >&2
+    exit 1
+fi
+
+# Create usage file directory if needed
+mkdir -p "$(dirname "$USAGE_FILE")"
+
+# Check if user wants raw output (non-interactive mode)
+if [[ "$*" == *"--output-format"* ]]; then
+    # User specified format, run as-is
+    exec "$REAL_GEMINI" "$@"
+fi
+
+# Run gemini with stream-json and process output
+"$REAL_GEMINI" --output-format stream-json "$@" 2>&1 | while IFS= read -r line; do
+    # Check if line is JSON
+    if echo "$line" | grep -q '^{'; then
+        # Parse the JSON event type
+        event_type=$(echo "$line" | python3 -c "import sys,json; d=json.loads(sys.stdin.read()); print(d.get('type',''))" 2>/dev/null)
+        
+        case "$event_type" in
+            "init")
+                # Extract model info
+                model=$(echo "$line" | python3 -c "import sys,json; d=json.loads(sys.stdin.read()); print(d.get('model','unknown'))" 2>/dev/null)
+                session_id=$(echo "$line" | python3 -c "import sys,json; d=json.loads(sys.stdin.read()); print(d.get('session_id',''))" 2>/dev/null)
+                ;;
+            "message")
+                # Display user/assistant messages
+                role=$(echo "$line" | python3 -c "import sys,json; d=json.loads(sys.stdin.read()); print(d.get('role',''))" 2>/dev/null)
+                content=$(echo "$line" | python3 -c "import sys,json; d=json.loads(sys.stdin.read()); print(d.get('content',''))" 2>/dev/null)
+                if [ "$role" = "assistant" ]; then
+                    echo -n "$content"
+                fi
+                ;;
+            "result")
+                # This is the usage data - log it!
+                timestamp=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
+                stats=$(echo "$line" | python3 -c "
+import sys, json
+d = json.loads(sys.stdin.read())
+stats = d.get('stats', {})
+print(json.dumps({
+    'timestamp': '$timestamp',
+    'model': '${model:-unknown}',
+    'input_tokens': stats.get('input_tokens', stats.get('input', 0)),
+    'output_tokens': stats.get('output_tokens', stats.get('output', 0)),
+    'total_tokens': stats.get('total_tokens', 0),
+    'cached_tokens': stats.get('cached', 0),
+    'duration_ms': stats.get('duration_ms', 0),
+    'tool_calls': stats.get('tool_calls', 0)
+}))
+" 2>/dev/null)
+                
+                if [ -n "$stats" ]; then
+                    echo "$stats" >> "$USAGE_FILE"
+                fi
+                echo "" # Final newline after response
+                ;;
+        esac
+    else
+        # Non-JSON output (errors, loading messages, etc.)
+        echo "$line" >&2
+    fi
+done