Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Set up Sentinel production error capture in your own Convex deployment.
.claude/skills/get-convex-convex-sentinel/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -42% | 0% |
<!-- GENERATED from convex-agents content/capabilities/sentinel.json — do not edit by hand. -->
Install @convex-dev/sentinel to capture production errors (server function failures, client JS/React crashes, OCC and scale signals) into a table in the user's OWN deployment, redacted at write time, then react to new ones. Data never leaves the user's deployment.
app.use(sentinel) in convex/convex.config.ts.window.onerror/unhandledrejection and breadcrumbs.convex data, run-once-query); react to new ones via the monitor's prod_error event.triage classifies each error and, for recurring non-transient ones, hands it to ai-runner to open a fix PR.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→fail | 16,761 | 12,871 | -23% | 1 | 1 | 0% | 2,350 | 1,984 | -16% | 0 | 0 | — |
case-06 | fail→fail | 20,969 | 16,740 | -20% | 1 | 1 | 0% | 2,838 | 2,878 | +1% | 0 | 0 | — |
case-01 | fail→pass | 21,462 | 11,287 | -47% | 1 | 1 | 0% | 4,411 | 2,636 | -40% | 0 | 0 | — |
case-02 | fail→pass | 24,414 | 10,141 | -58% | 1 | 1 | 0% | 3,563 | 2,218 | -38% | 0 | 0 | — |
case-03 | fail→pass | 18,079 | 9,866 | -45% | 1 | 1 | 0% | 2,557 | 2,186 | -15% | 0 | 0 | — |
case-04 | pass→fail | 16,173 | 9,877 | -39% | 1 | 1 | 0% | 2,422 | 1,995 | -18% | 0 | 0 | — |
case-07 | fail→pass | 13,718 | 3,436 | -75% | 1 | 1 | 0% | 2,256 | 961 | -57% | 0 | 0 | — |
case-08 | fail→pass | 16,046 | 7,870 | -51% | 1 | 1 | 0% | 2,595 | 1,500 | -42% | 0 | 0 | — |
case-09 | pass→pass | 26,739 | 11,733 | -56% | 1 | 1 | 0% | 4,986 | 2,414 | -52% | 0 | 0 | — |
case-10 | pass→pass | 19,980 | 9,103 | -54% | 1 | 1 | 0% | 3,219 | 1,913 | -41% | 0 | 0 | — |
case-11 | pass→pass | 15,144 | 12,886 | -15% | 1 | 1 | 0% | 2,294 | 2,058 | -10% | 0 | 0 | — |
case-12 | fail→pass | 19,109 | 10,658 | -44% | 1 | 1 | 0% | 2,447 | 2,024 | -17% | 0 | 0 | — |
case-13 | fail→pass | 14,893 | 2,598 | -83% | 1 | 1 | 0% | 2,059 | 625 | -70% | 0 | 0 | — |
case-14 | fail→pass | 16,614 | 5,703 | -66% | 1 | 1 | 0% | 2,195 | 1,059 | -52% | 0 | 0 | — |
case-15 | pass→pass | 19,530 | 9,568 | -51% | 1 | 1 | 0% | 3,582 | 2,024 | -43% | 0 | 0 | — |
case-16 | fail→pass | 15,103 | 5,925 | -61% | 1 | 1 | 0% | 2,393 | 1,218 | -49% | 0 | 0 | — |
case-17 | pass→pass | 23,872 | 11,073 | -54% | 1 | 1 | 0% | 3,114 | 1,799 | -42% | 0 | 0 | — |
case-18 | fail→pass | 12,470 | 2,367 | -81% | 1 | 1 | 0% | 1,980 | 701 | -65% | 0 | 0 | — |
case-19 | pass→pass | 20,587 | 9,587 | -53% | 1 | 1 | 0% | 3,119 | 1,716 | -45% | 0 | 0 | — |
case-20 | pass→pass | 17,521 | 9,089 | -48% | 1 | 1 | 0% | 2,263 | 1,230 | -46% | 0 | 0 | — |
case-21 | fail→pass | 14,571 | 3,949 | -73% | 1 | 1 | 0% | 1,837 | 930 | -49% | 0 | 0 | — |
case-22 | pass→pass | 9,025 | 2,426 | -73% | 1 | 1 | 0% | 1,344 | 662 | -51% | 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. The headline lift of +41 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are 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.