Fix audit logging#27
Merged
vishalveerareddy123 merged 4 commits intoFast-Editor:mainfrom Jan 31, 2026
Merged
Conversation
Problem: PR Fast-Editor#20 introduced new code but unintentionally introduced inconsistent code. Changes implemented: 1. Files cleaned up 2. Missing dns-logger.js added 3. New code added Testing: success: npm run test:unit success: Claude connects to Lynkr correctly
Collaborator
|
Thanks @MichaelAnders I will take a look at this one today |
Integrates the audit logging infrastructure with the orchestrator's runAgentLoop function to capture all LLM request/response pairs for compliance and debugging purposes. Changes: - Add crypto module import for correlation ID generation - Add getDestinationUrl() helper to resolve provider endpoints - Instantiate audit logger at the start of runAgentLoop - Log LLM requests before invokeModel() with correlation IDs - Log LLM responses after invokeModel() with token usage and latency - Handle streaming, success, and error response cases - Add logs/ directory to .gitignore The audit logger is a no-op when disabled, ensuring zero overhead when not in use. All logs use structured JSON format with correlation IDs to link request/response pairs. Implements plan from PR Fast-Editor#27 for intelligent audit logging.
Contributor
Author
|
I have added the missing code as my plans for the next 3 days have changed this morning. Testing: |
Contributor
Author
I have another PR coming up based on these changes (Ollama specific and "correct" this time). I'll create the new PR once you've accepted this one, I don't want to mix this - it'll get too big and complicated. IGNORE 9ae1d30 - it is a local branch, WIP |
MichaelAnders
added a commit
to MichaelAnders/Lynkr
that referenced
this pull request
Jan 28, 2026
This PR builds on PR Fast-Editor#27's audit logging infrastructure to fix critical issues when using Ollama as the LLM provider. ## Problems Fixed ### 1. Ollama JSON String Tool Call Parsing **Problem**: Ollama returns tool calls as JSON strings in the content field instead of structured tool_calls arrays, causing malformed displays and execution failures. **Example**: `{"name": "Bash", "parameters": {"command": "ls"}}` displayed as raw JSON instead of executing. **Solution**: Enhanced ollama-utils.js to detect and parse JSON strings, converting them to proper Anthropic tool_use format. ### 2. File Path Extraction Endless Loop **Problem**: Commands like "ls" created an endless loop. The CLI sends two parallel requests: the main command AND a file path extraction request expecting XML format. Ollama kept returning Grep tool calls instead of the requested `<filepaths></filepaths>` XML, creating an infinite loop. **Solution**: Added detection for file path extraction requests and strip tools from the payload, forcing Ollama to return plain XML text. ### 3. Enhanced Audit Logging Builds on PR Fast-Editor#27's fixes for: - **Missing Response Content**: Ollama's `message.message` format now captured - **Model Name Mismatch**: Both requested and actual model logged correctly - **User Messages Buried**: Actual user input extracted from wrapped content **New enhancements**: - Query-response pairing with correlation IDs for easy audit trail - Application log mirroring to audit file for debugging (configurable levels) - Oversized error logging integration ## Changes **Core Fixes**: - src/clients/ollama-utils.js: Add JSON string detection and parsing - New: isToolCallJsonString() helper - New: parseToolCallFromJsonString() helper - Enhanced: buildAnthropicResponseFromOllama() for centralized conversion - src/orchestrator/index.js: Add file path extraction detection - Detects XML format requests from CLI - Strips tools to force text-only responses - Prevents endless tool call loops **Audit Logging Enhancements**: - src/logger/audit-logger.js: Add query-response pairing - New: logQueryResponsePair() method - Correlates requests with responses using correlation IDs - src/logger/index.js: Add application log mirroring - Mirrors app logs to audit file (configurable: none/error/warn/info/debug) - Provides comprehensive debugging visibility - src/config/index.js: Add appLogLevel configuration - .env.example: Document all audit logging options (+51 lines) **Testing**: - test/ollama-json-string-parsing.test.js: 10 comprehensive tests (all passing) ## Configuration New environment variables: ```bash # Mirror application logs to audit file for debugging LLM_AUDIT_APP_LOG_LEVEL=info # none|error|warn|info|debug ``` ## Verification 1. Start Lynkr: `~/start_lynkr` 2. Test ls command: Works without endless loop 3. Test tool calls: Properly parsed and executed 4. Check audit logs: `tail -50 logs/llm-audit.log | jq` Builds on PR Fast-Editor#27 audit logging infrastructure.
Collaborator
|
Thank you @MichaelAnders |
After adding more comprehensive logging to track context flow through the request pipeline, discovered a bug in Ollama system prompt handling. BUG FIX: Ollama System Prompt Handling - Problem: System prompts were being flattened into the first user message for Ollama - Impact: Models received instructions as conversational context instead of system directives, causing meta-commentary responses instead of tool execution - Fix: Keep system prompt separate for Ollama (same pattern as other providers) - Files: src/orchestrator/index.js, src/clients/databricks.js Diagnostic Logging Added: - MEMORY_DEBUG: Track memory retrieval, formatting, and injection (src/memory/retriever.js) - SESSION_DEBUG: Track session reuse vs creation, age, and history size (src/sessions/store.js) - CONTEXT_FLOW: Track system prompt transformations through the pipeline (src/orchestrator/index.js) The diagnostic logging revealed how system prompts flow through the request pipeline and helped identify where the flattening was occurring.
MichaelAnders
added a commit
to MichaelAnders/Lynkr
that referenced
this pull request
Jan 30, 2026
Collaborator
|
Thank you Micheal for all your contributions to the Lynkr Project |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR addresses issues with the audit logging system and fixes a critical bug in Ollama system prompt handling that was causing models to produce meta-commentary instead of executing commands.
Changes
1. Fix Audit Logging Infrastructure (bd9c1a6)
2. Integrate Audit Logging with Orchestrator (e69d6f4)
3. Fix Ollama System Prompt Handling (8b2b7cf) 🐛
Critical Bug Fix: System prompts were being flattened into user message content for Ollama, causing models to see instructions as conversational context instead of system directives.
Impact: Models produced meta-commentary like "It seems like you're ready to start a conversation..." instead of executing commands like "ls" or "pwd".
Fix: Keep system prompt separate as first message with
role: "system", consistent with all other providers (OpenRouter, Azure OpenAI, llama.cpp, LM Studio).Diagnostic Logging Added:
MEMORY_DEBUG: Track memory retrieval, formatting, and injectionSESSION_DEBUG: Track session reuse vs creation, age, and history sizeCONTEXT_FLOW: Track system prompt transformations through the pipeline5 files changed, +344/-28 lines
Testing
npm run test:unitFiles Changed
Total: 18 files, ~3,444 additions, ~55 deletions
Key files:
src/orchestrator/index.js- Audit integration + Ollama system prompt fixsrc/clients/databricks.js- Ollama system prompt handlingsrc/logger/audit-logger.js- New audit logging infrastructuresrc/logger/deduplicator.js- Deduplication for audit logssrc/memory/retriever.js- Memory debug loggingsrc/sessions/store.js- Session debug loggingdocs/memory-stale-context-fix.md- Comprehensive documentationRelated Issues
Fixes system prompt flattening causing meta-commentary responses from Ollama models.