Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Review many client Odoo databases at once through odoo-mcp's cross-instance tools — fleet-wide accounting health, per-client aging, partial-failure triage — for agencies and partners managing 5–50 instances. Use when the user asks "which client...", "across all instances/databases", or wants a fleet/portfolio status.
.claude/skills/erpipe-org-odoo-agency-fleet-review/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -26% | 0% |
You are answering questions across a fleet of client Odoo databases through one odoo-mcp server with named instances. Every result is tagged with its _instance; one client being down must never sink the whole answer.
ODOO_CONFIG_FILE with an instances map).list_instances shows names, tags, and which allow cross-instance reads ("cross_instance": false opts a client out — respect it silently).
through the normal gate, one client at a time.
list_instances — report count, tags, default, andany opted-out clients (just the count, not a complaint).
accounting_health_across_instances(instances="all")(or {"tags": ["managed"]}). For fleets >10, run via submit_async_task and poll.
errors — an unreachable client is a finding in itself (report it, with diagnose_odoo_call output if the human wants the cause), not a reason to retry the whole fan-out.
the human asked about (e.g. overdue AR). For the worst clients, drill down with instance-scoped calls: receivable_payable_aging(instance="client_x"), search_records(..., instance="client_x").
merged across instances (different currencies/configs) — compare counts and per-client aggregates, never invent a fleet-wide average.
instance and goes through the write gate there. Approval tokens are instance-bound; never reuse one across clients.
Fleet summary (reachable/unreachable/opted-out counts), the ranked per-client table with _instance labels, drill-down findings, and a follow-up list grouped by client.
response says redacted_fields, that is policy, not an error.
ODOO_MCP_CROSS_INSTANCE_WORKERS); for very largefleets prefer the async path over repeated synchronous sweeps.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | pass→pass | 9,088 | 4,988 | -45% | 1 | 1 | 0% | 1,396 | 1,472 | +5% | 0 | 0 | — |
case-01 | fail→fail | 27,669 | 2,668 | -90% | 1 | 1 | 0% | 5,119 | 807 | -84% | 0 | 0 | — |
case-02 | fail→fail | 16,497 | 3,031 | -82% | 1 | 1 | 0% | 2,655 | 828 | -69% | 0 | 0 | — |
case-03 | fail→fail | 20,363 | 3,597 | -82% | 1 | 1 | 0% | 3,451 | 804 | -77% | 0 | 0 | — |
case-04 | fail→pass | 13,051 | 4,812 | -63% | 1 | 1 | 0% | 1,942 | 1,377 | -29% | 0 | 0 | — |
case-05 | fail→fail | 4,286 | 4,638 | +8% | 1 | 1 | 0% | 640 | 831 | +30% | 0 | 0 | — |
case-07 | fail→pass | 8,754 | 2,635 | -70% | 1 | 1 | 0% | 1,338 | 1,015 | -24% | 0 | 0 | — |
case-08 | pass→fail | 9,070 | 4,194 | -54% | 1 | 1 | 0% | 1,249 | 996 | -20% | 0 | 0 | — |
case-09 | pass→pass | 8,048 | 4,009 | -50% | 1 | 1 | 0% | 1,246 | 1,203 | -3% | 0 | 0 | — |
case-10 | pass→pass | 6,620 | 2,944 | -56% | 1 | 1 | 0% | 1,026 | 1,076 | +5% | 0 | 0 | — |
case-11 | fail→pass | 12,242 | 4,672 | -62% | 1 | 1 | 0% | 1,839 | 1,356 | -26% | 0 | 0 | — |
case-12 | fail→fail | 13,170 | 2,120 | -84% | 1 | 1 | 0% | 1,868 | 929 | -50% | 0 | 0 | — |
case-13 | fail→pass | 7,756 | 2,285 | -71% | 1 | 1 | 0% | 1,224 | 1,011 | -17% | 0 | 0 | — |
case-14 | fail→pass | 11,292 | 4,408 | -61% | 1 | 1 | 0% | 1,803 | 1,335 | -26% | 0 | 0 | — |
case-15 | fail→pass | 13,450 | 3,087 | -77% | 1 | 1 | 0% | 2,121 | 1,131 | -47% | 0 | 0 | — |
case-16 | pass→pass | 4,930 | 3,669 | -26% | 1 | 1 | 0% | 690 | 1,132 | +64% | 0 | 0 | — |
case-17 | fail→pass | 13,375 | 1,432 | -89% | 1 | 1 | 0% | 2,085 | 824 | -60% | 0 | 0 | — |
case-18 | fail→pass | 14,011 | 4,940 | -65% | 1 | 1 | 0% | 2,159 | 1,392 | -36% | 0 | 0 | — |
case-19 | fail→pass | 8,289 | 1,979 | -76% | 1 | 1 | 0% | 1,242 | 911 | -27% | 0 | 0 | — |
case-20 | pass→pass | 10,301 | 8,178 | -21% | 1 | 1 | 0% | 1,813 | 1,982 | +9% | 0 | 0 | — |
case-21 | pass→pass | 11,040 | 9,047 | -18% | 1 | 1 | 0% | 1,870 | 2,144 | +15% | 0 | 0 | — |
case-22 | pass→pass | 10,775 | 6,458 | -40% | 1 | 1 | 0% | 1,867 | 1,735 | -7% | 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. 22 cases were attempted, and 17 counted toward the lift figure. The other 5 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 +36 percentage points is the difference between those two pass rates over the 17 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.