Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Audit an Odoo database's data quality with evidence before trusting AI answers, importing, or migrating — duplicates, missing required values, orphaned references, format anomalies — and drive remediation through odoo-mcp's gated write workflow. Use when the user asks to "check data quality", "clean up data", "prepare for migration", "find duplicates", or when aggregate answers look suspicious.
.claude/skills/erpipe-org-odoo-data-quality-gate/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 11 |
| gemini-3.1-pro-preview | 100% | 3 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -43% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -13% | 0% |
You are running a data-quality audit against a live Odoo database through the odoo-mcp server (tools named data_quality_report, diagnose_access, preview_write, …). Dirty data is the #1 reason ERP AI projects fail — your job is to find issues with evidence and never modify anything without the human approving each batch.
health_check).ODOO_MCP_ENABLE_WRITES=1 —remediation proposals are still valuable without it.
audit: res.partner, product.template, account.move. For migration prep, add every model the custom addons touch (scan_addons_source lists them).
data_quality_report(model=...). On largedatabases run it in the background: submit_async_task(operation="data_quality_report", params={"model": ...}) then poll get_async_task.
summary.checks_with_issues and show evidence. Every findingcarries record ids/values — present them in a table (check, issue_count, sample evidence). Never summarize away the ids; the human needs them.
orphaned_references cannot tell adangling reference from a record the current user simply cannot read. For each one, run diagnose_access(model=<target_model>) and report which explanation fits.
duplicates, fill required fields, archive orphans) into small batches of explicit record ids with the exact new values.
preview_write → show the diff → validate_write → human confirms → execute_approved_write(confirm=true). Never call execute_method for writes; it is blocked by design.
counts.
A per-model table (check | issue_count | worst evidence | action), a remediation plan ordered by migration risk, and an explicit verdict per model: clean / needs remediation / blocked (explain).
human's explicit confirmation for that batch.
redacted_fields in responses — never ask the user to lift thefield ACL to "see more".
summary.checks_errored), say so — do not present apartial audit as complete.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 10,038 | 5,115 | -49% | 1 | 1 | 0% | 1,711 | 1,511 | -12% | 0 | 0 | — |
case-06 | fail→pass | 20,888 | 7,701 | -63% | 1 | 1 | 0% | 3,652 | 2,071 | -43% | 0 | 0 | — |
case-01 | fail→fail | 28,380 | 4,131 | -85% | 1 | 1 | 0% | 4,742 | 1,043 | -78% | 0 | 0 | — |
case-02 | fail→fail | 22,661 | 4,426 | -80% | 1 | 1 | 0% | 3,827 | 995 | -74% | 0 | 0 | — |
case-03 | fail→fail | 36,507 | 3,849 | -89% | 1 | 1 | 0% | 6,231 | 966 | -84% | 0 | 0 | — |
case-04 | pass→pass | 15,900 | 9,324 | -41% | 1 | 1 | 0% | 2,590 | 1,778 | -31% | 0 | 0 | — |
case-07 | fail→pass | 12,457 | 4,705 | -62% | 1 | 1 | 0% | 1,783 | 1,380 | -23% | 0 | 0 | — |
case-08 | fail→pass | 15,334 | 7,642 | -50% | 1 | 1 | 0% | 2,371 | 1,841 | -22% | 0 | 0 | — |
case-09 | fail→fail | 12,583 | 6,680 | -47% | 1 | 1 | 0% | 1,872 | 1,731 | -8% | 0 | 0 | — |
case-10 | fail→pass | 12,710 | 7,078 | -44% | 1 | 1 | 0% | 2,086 | 1,812 | -13% | 0 | 0 | — |
case-11 | fail→pass | 13,129 | 4,693 | -64% | 1 | 1 | 0% | 1,966 | 1,458 | -26% | 0 | 0 | — |
case-12 | fail→fail | 6,926 | 2,369 | -66% | 1 | 1 | 0% | 966 | 1,060 | +10% | 0 | 0 | — |
case-13 | fail→pass | 12,192 | 3,311 | -73% | 1 | 1 | 0% | 1,972 | 1,232 | -38% | 0 | 0 | — |
case-14 | fail→fail | 14,901 | 5,143 | -65% | 1 | 1 | 0% | 2,344 | 1,523 | -35% | 0 | 0 | — |
case-15 | pass→pass | 11,094 | 4,570 | -59% | 1 | 1 | 0% | 1,696 | 1,352 | -20% | 0 | 0 | — |
case-16 | pass→pass | 11,208 | 4,301 | -62% | 1 | 1 | 0% | 1,678 | 1,300 | -23% | 0 | 0 | — |
case-17 | fail→fail | 7,816 | 1,688 | -78% | 1 | 1 | 0% | 1,358 | 882 | -35% | 0 | 0 | — |
case-18 | fail→pass | 10,478 | 4,770 | -54% | 1 | 1 | 0% | 1,902 | 1,507 | -21% | 0 | 0 | — |
case-19 | pass→pass | 16,612 | 10,846 | -35% | 1 | 1 | 0% | 2,454 | 2,383 | -3% | 0 | 0 | — |
case-20 | fail→pass | 6,167 | 1,834 | -70% | 1 | 1 | 0% | 1,047 | 995 | -5% | 0 | 0 | — |
case-21 | pass→pass | 7,722 | 6,036 | -22% | 1 | 1 | 0% | 1,577 | 1,854 | +18% | 0 | 0 | — |
case-22 | pass→pass | 19,577 | 16,056 | -18% | 1 | 1 | 0% | 3,367 | 3,528 | +5% | 0 | 0 | — |
case-23 | pass→pass | 14,740 | 13,603 | -8% | 1 | 1 | 0% | 2,963 | 3,340 | +13% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 20 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +39 percentage points is the difference between those two pass rates over the 20 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.