Extract Log Debug

Extract and analyze LibrAgent debug logs for troubleshooting and analysis.

Quick Start

Extract logs using the Python script:

# Extract last 100 lines
python scripts/extract_logs.py -n 100

# Extract all errors with 5 lines of context
python scripts/extract_logs.py --pattern "[ERROR]" --context 5

# Extract planning logs
python scripts/extract_logs.py --pattern "PLANNING" -n 5000

# Save to custom file
python scripts/extract_logs.py --pattern "[WARN]" -o warnings.txt

Log File Location

LibrAgent stores logs in platform-specific directories:

• Windows: %LOCALAPPDATA%\com.fritzprix.libragent\logs\libragent.log
• macOS: ~/Library/Logs/com.fritzprix.libragent/libragent.log
• Linux: ~/.local/share/com.fritzprix.libragent/logs/libragent.log

The script automatically detects and uses the correct path.

Common Patterns

Error Extraction

python scripts/extract_logs.py --pattern "[ERROR]" --context 10

Component-Specific Logs

# Agent workflow logs
python scripts/extract_logs.py --pattern "agent_" -n 5000

# MCP tool execution
python scripts/extract_logs.py --pattern "MCPServiceProxy" --context 10

# Planning operations
python scripts/extract_logs.py --pattern "PLANNING" --context 5

# Browser automation
python scripts/extract_logs.py --pattern "BrowserServer" --context 10

Recent Activity

# Last 500 lines for quick check
python scripts/extract_logs.py -n 500

# Last 5000 lines for detailed analysis
python scripts/extract_logs.py -n 5000

Pattern Reference

For comprehensive list of log patterns and search strategies, see log_patterns.md.

Key pattern categories:

• Error patterns ([ERROR], Failed to, panic)
• Component logs (agent_, MCPServiceProxy, BrowserServer)
• Workflow phases (Think phase, Act phase, Observe phase)
• Tool operations (call_tool, list_tools)
• Performance indicators (Duration:, elapsed)

Debugging Workflows

1. Investigate Error

# Extract errors
python scripts/extract_logs.py --pattern "[ERROR]" --context 10 -o errors.txt

# Review errors.txt for:
# - Error message and stack trace
# - Preceding operations (context lines above)
# - Affected component (agent, MCP, builtin server)
# - Session ID for tracking

2. Analyze Agent Workflow

# Extract workflow logs
python scripts/extract_logs.py --pattern "agent_" -n 5000 -o workflow.txt

# Review workflow.txt for:
# - Session lifecycle (create, start, stop)
# - Workflow phases (Think, Act, Observe)
# - Tool execution sequences
# - Loop iterations and completion

3. Debug Tool Execution

# Extract tool logs
python scripts/extract_logs.py --pattern "call_tool" --context 10 -o tools.txt

# Review tools.txt for:
# - Tool name and arguments
# - Routing (builtin vs external)
# - Execution results or errors
# - Response structure

4. Track Session Activity

# Extract specific session logs (replace <session-id> with actual ID)
python scripts/extract_logs.py --pattern "<session-id>" --context 5 -o session.txt

# Review session.txt for:
# - Session initialization
# - Tool calls made during session
# - Workflow status changes
# - Errors specific to session

Output Format

Pattern Match Output

When using --pattern, output includes:

• Match count and line ranges
• Line numbers with markers (> for matching lines)
• Context lines before and after matches
• Merged ranges for adjacent matches

Example:

=== Match 1: Lines 1234-1244 ===
  1234   [DEBUG] Starting operation
  1235 > [ERROR] Failed to execute tool
  1236   Stack trace: ...
  1237   at module.rs:123

Full Line Output

Without --pattern, outputs raw log lines preserving original format.

Tips

• Use wider context (10-20 lines) for workflow analysis
• Use narrower context (3-5 lines) for quick error checks
• Combine -n with --pattern to search recent logs only
• Check log_patterns.md for component-specific patterns
• For performance issues, search for "Duration:" or "elapsed" patterns