Conversion issues
Non-interactive child. Do not ask questions. Call configure with project_dir from the parent, then run the command from that project.
The first command is not the result. It collects issues and leaves problem_description and recommended_fix empty. Returning after that command, or checking only counts and files, is a failure. Done means the regenerated JSON has both fields filled for every summary row, or there are zero pairs.
1. Collect
scai assessment conversion-issues
This reads the code unit registry and converted files. It writes every distinct pair. Do not limit the run to a subset. It writes:
<project_dir>/artifacts/assessment/conversion-issues.json<project_dir>/artifacts/assessment/conversion-issues/conversion_issues.csv<project_dir>/artifacts/assessment/conversion-issues/conversion_issues.zip
Zero SSC-EWI- pairs is success: JSON only, no CSV, no ZIP, and no enrichment. Stop there and return ok. ASM0045 (no registry) is error, not skipped.
2. Enrich
Required whenever the summary panel has rows. Read panels where id is summary. For each row, take code and description only.
For each distinct pair, write a Problem Description and a Recommended Fix that name the construct in the description. Use sources in this order:
- AIM conversion-issue documentation in the Snowflake migrations documentation
- The source dialect documentation
- Snowflake SQL documentation
The fix must name a concrete replacement. Do not leave <!-- ewi-skill:agent-must-fill -->. Do not add an AI disclaimer or the heading "AI-generated recommendation — not a mandatory fix". The CSV places that warning in a Jira {warning} panel above the fix. The report and dashboard show only the fix, under a heading that says it is AI-generated and must be validated.
Write <project_dir>/artifacts/assessment/conversion-issues/updates.json:
[{ "code": "SSC-EWI-0040", "description": "THE 'SET XACT_ABORT' CLAUSE IS NOT SUPPORTED IN SNOWFLAKE",
"problem_description": "...", "recommended_fix": "..." }]
When updates.json already exists, keep its entries and research only the rows whose problem_description is still empty.
3. Regenerate
scai assessment conversion-issues --updates artifacts/assessment/conversion-issues/updates.json
Run this after writing updates.json, before the HTML report. Then read conversion-issues.json again. Every summary row must have a non-empty problem_description and recommended_fix. If any field is empty, fill it and regenerate. Do not return ok while any summary field is empty. The parent does not upload the ZIP. The report shows the ZIP path and the Share issues link.
Return
JSON only:
{ "sub_skill": "conversion-issues", "status": "ok", "output_json": "<abs>/artifacts/assessment/conversion-issues.json",
"summary": "<distinct pairs> pairs, <instances> instances", "error": null }
Any other failure: status "error", output_json null, error "<message>".