crm-export-auditorlisted

Anthropic's [sales plugin](https://github.com/anthropics/knowledge-work-plugins) ships `pipeline-review` and `forecast`, and both grade what is in the export. Neither tells you the export is unreliable. This skill does only that: it treats the CSV as the thing under test, not the pipeline, and reports the duplicates, gaps, contradictions and format drift that quietly move a forecast number. It needs no database access and no tooling — paste the header with rows, or attach the file — so it runs the same in claude.ai, Claude Code, and Claude Cowork. [/audit-crm](/commands/sales/audit-crm) is the runner for larger files. ## When to use this skill - The forecast number and the CRM number disagree and nobody knows which is wrong. - Before a CRM migration, a merge of two instances, or a dedupe run. - After an import, an enrichment job, or a sequencer syncing contacts back in. - A rep left and their records need reassignment. - Before handing pipeline data to a model, a dashboard, or a board deck. - Quarterly, as the hygiene pass that runs before pipeline review rather than during it. > [!NOTE] > This is read-only by design. It produces a change list; it never edits the CRM and never rewrites your file. Export only the columns the audit needs — a full export of every contact's email and phone is more personal data than this job requires, and [Data privacy for LLM apps](/guides/ai-safety/data-privacy-for-llm-apps) covers why that matters. Every threshold below is a default you can