[Feature]: Improve logging infrastructure and add LLM prompt logging

[deepakn97]

Feature Summary

Improve the PAS logging infrastructure to be more useful for debugging and development. This includes:

1. Adding LLM prompt logging to capture actual context sent to agents
2. Reducing noise in debug-level logs
3. Adding more informative log messages at appropriate levels

Motivation and Use Case

Current problems:

• Debug logs are very noisy and don't relay useful information for debugging agent behavior
• No way to inspect actual LLM prompts during scenario execution - existing logs (agent.log, pas.log) show activity but not prompt content
• Hard to trace what context an agent sees when making decisions
• Difficult to verify fixes like duplicate dynamic log filtering (Issue [Bug]: Duplicate dynamic logs in agent history polluting LLM context #33)

Use cases:

• Debugging agent decision-making by inspecting actual prompts
• Verifying context management (e.g., dynamic log filtering) is working correctly
• Reducing log noise to focus on meaningful events
• Research analysis of agent context and behavior

Proposed Solution

1. LLM Prompt Logging

Integration point: Override append_to_world_logs() in StateAwareEnvironmentWrapper (pas/environment.py).

The log_callback passed to agents is set to env.append_to_world_logs in scenario_runner.py. When the base agent's step() function creates an LLMInputLog, we can intercept it and write prompt content to a dedicated prompts log file.

Key files:

• pas/environment.py:StateAwareEnvironmentWrapper - add override method
• pas/scenario_runner.py:202,214 - where log_callback is set

2. Log Level Improvements

• Review current DEBUG statements and promote informative ones to INFO
• Demote noisy DEBUG statements to TRACE level (or remove)
• Add structured logging for key events (agent turns, proposals, tool calls)

3. Log Format Improvements

• Consider structured logging format for easier parsing
• Add agent ID and turn number context to log messages
• Separate prompt logs into dedicated file

Alternatives Considered

1. Keep current logging as-is - Not viable as current logs are hard to use for debugging
2. External logging service - Overkill for a research framework

Component

Core

Priority

Medium - Nice to have

Additional Context

This feature was identified while manually testing Issue #33 (duplicate dynamic logs fix). Without prompt logging, we cannot verify the fix is working in production runs. The noisy debug logs also made it difficult to find relevant information.