Summary

The x2a-convertor migrate command enters an infinite loop during Ansible code generation when it encounters ansible-lint issues it cannot resolve. The LLM agent repeatedly rewrites the same file and re-runs ansible-lint, never succeeding, until it hits the LangGraph recursion limit of 500.

Steps to Reproduce

1. Clone the x2ansible chef-examples repo:

   git clone https://github.com/x2ansible/chef-examples.git
   cd chef-examples
   

2. Run init and analyze for the nginx-multisite cookbook
3. Run the migrate command:

   podman run --rm -ti \
     -v $(pwd)/:/app/source:Z \
     -e LLM_MODEL=anthropic.claude-3-7-sonnet-20250219-v1:0 \
     -e AWS_REGION=eu-west-2 \
     -e AWS_BEARER_TOKEN_BEDROCK=$AWS_BEARER_TOKEN_BEDROCK \
     quay.io/x2ansible/x2a-convertor:latest \
     migrate --source-dir /app/source/ --source-technology Chef \
     --high-level-migration-plan migration-plan.md \
     --module-migration-plan migration-plan-nginx-multisite.md \
     "Convert the 'nginx-multisite' module"
   

Expected Behavior

The migrate agent should detect when it has failed to fix the same ansible-lint issues after a reasonable number of retries (e.g., 3-5 attempts) and either:

• Accept the code with known lint warnings and proceed
• Report the unfixable issues and move on to the next task
• Bail out gracefully with a clear error message

Actual Behavior

The agent gets stuck in a loop on ansible/roles/nginx_multisite/tasks/security.yml:

1. Generates/rewrites the file using ansible_write tool
2. Runs ansible_lint tool - finds 2 issues
3. Rewrites the file to try to fix the issues
4. Runs ansible_lint again - still finds 2 issues
5. Repeats this cycle hundreds of times
6. Eventually hits the LangGraph recursion limit of 500:

   Error: Recursion limit of 500 reached without hitting a stop condition.
   For troubleshooting, visit: https://docs.langchain.com/oss/python/langgraph/errors/GRAPH_RECURSION_LIMIT
   

Additionally, ansible-lint itself is throwing internal errors during each iteration:

ERROR:ansiblelint.skip_utils: Error trying to append skipped rules
RuntimeError: Error in matching skip comment to a task

This suggests the lint issue may be caused by an ansible-lint bug with skip comments, making it impossible for the LLM to fix by rewriting the Ansible code.

Impact

• Wasted LLM API credits: Hundreds of Bedrock API calls for the same failing operation
• CI pipeline failure: The Jenkins job fails after running for an extended period
• No usable output: The migration does not complete, so no Ansible code is produced for the module

Environment

• x2a-convertor image: quay.io/x2ansible/x2a-convertor:latest
• LLM Model: anthropic.claude-3-7-sonnet-20250219-v1:0
• Source cookbook: nginx-multisite from https://github.com/x2ansible/chef-examples.git
• Jenkins job: flightpath-x2ansible build #7

Suggested Fix

Add a maximum retry counter for lint-fix cycles in the migrate agent. After N failed attempts to fix the same lint issues on the same file, the agent should move on rather than retrying indefinitely.