Use when Codex needs to process legal evidence files such as .mbox, .msg, and .pdf into review-ready exports, OCR quality reports, retry flows, or unified case text outputs; especially when the task requires workflow selection, safe execution, deterministic artifacts, or recovery guidance.
Route the request to the right CLI workflow first. Keep execution safe, explain the outcome clearly, and load only the reference file that matches the chosen workflow.
Use this table before proposing commands:
| User intent | Workflow |
|---|---|
| Export a raw mailbox for review or LLM ingestion | export-mbox |
| Process PDFs with OCR fallback and quality reporting | pdf-ingest |
| Retry only failed or weak OCR files from a prior report | pdf-retry |
Combine .msg and .pdf evidence into one deterministic text export | unified-export |
| Mixed or unsupported evidence request | narrow to supported types or reject cleanly |
Supported V1 evidence types:
.mbox.msg.pdfUnsupported in V1:
.docx.xlsxBefore running anything:
When reporting results:
When a run partially fails:
export-mbox
.mbox path.pdf-ingest
balanced unless the user has a clear reason to use quick or thorough.pdf-retry
quality_report.csv.unified-export
.msg and .pdf.path, date-signal, date-query.date-query mode, describe fallback behavior if Ollama fails.Load only what you need:
references/workflows.md for workflow selection, mailbox export patterns, shared summary language, and unsupported-input narrowing.references/pdf-ingest.md when the workflow is pdf-ingest or pdf-retry.references/unified-export.md when the workflow is unified-export.Do not load all references by default.