Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Advanced Observal operations including session reconciliation, CLI upgrades and downgrades, complete uninstallation, and local fallback mode for offline use. Use when the user wants to reconcile sessions, upgrade or downgrade the CLI, uninstall Observal, or write agent configs locally when the server is unreachable.
.claude/skills/observal-observal-advanced/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -49% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -9% | 0% |
--output json whenever supported. Parse list results from items and pagination fields.--help before acting when a path or flag is uncertain.Read Recovery workflows completely before executing.
Connection failed or Not configured, and only when the user still wants local files written.Report sessions discovered, queued, skipped, or failed for reconciliation; old and new versions for CLI changes; or exact local paths for fallback. Include unresolved warnings and the command needed once the server is reachable.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-13 | fail→pass | 22,125 | 5,080 | -77% | 1 | 1 | 0% | 886 | 1,124 | +27% | 0 | 0 | — |
case-01 | fail→fail | 9,484 | 6,711 | -29% | 1 | 1 | 0% | 1,502 | 624 | -58% | 0 | 0 | — |
case-02 | fail→fail | 6,007 | 5,348 | -11% | 1 | 1 | 0% | 225 | 599 | +166% | 0 | 0 | — |
case-03 | fail→fail | 11,095 | 7,426 | -33% | 1 | 1 | 0% | 1,732 | 745 | -57% | 0 | 0 | — |
case-04 | fail→fail | 5,037 | 7,200 | +43% | 1 | 1 | 0% | 606 | 805 | +33% | 0 | 0 | — |
case-05 | fail→fail | 7,800 | 9,077 | +16% | 1 | 1 | 0% | 311 | 666 | +114% | 0 | 0 | — |
case-06 | pass→fail | 9,979 | 7,145 | -28% | 1 | 1 | 0% | 984 | 897 | -9% | 0 | 0 | — |
case-07 | fail→fail | 7,521 | 5,397 | -28% | 1 | 1 | 0% | 1,047 | 580 | -45% | 0 | 0 | — |
case-08 | pass→fail | 11,262 | 9,447 | -16% | 1 | 1 | 0% | 1,817 | 619 | -66% | 0 | 0 | — |
case-09 | fail→fail | 9,065 | 9,541 | +5% | 1 | 1 | 0% | 1,352 | 841 | -38% | 0 | 0 | — |
case-10 | pass→fail | 6,808 | 7,963 | +17% | 1 | 1 | 0% | 813 | 834 | +3% | 0 | 0 | — |
case-11 | fail→fail | 5,402 | 6,946 | +29% | 1 | 1 | 0% | 721 | 760 | +5% | 0 | 0 | — |
case-12 | pass→fail | 10,505 | 7,918 | -25% | 1 | 1 | 0% | 1,709 | 769 | -55% | 0 | 0 | — |
case-14 | fail→pass | 14,759 | 13,749 | -7% | 1 | 1 | 0% | 2,218 | 1,867 | -16% | 0 | 0 | — |
case-15 | fail→pass | 10,289 | 6,159 | -40% | 1 | 1 | 0% | 1,795 | 1,303 | -27% | 0 | 0 | — |
case-16 | fail→pass | 23,521 | 6,306 | -73% | 1 | 1 | 0% | 2,567 | 1,301 | -49% | 0 | 0 | — |
case-17 | pass→pass | 9,476 | 5,035 | -47% | 1 | 1 | 0% | 1,440 | 1,195 | -17% | 0 | 0 | — |
case-18 | pass→pass | 11,682 | 5,252 | -55% | 1 | 1 | 0% | 1,714 | 1,190 | -31% | 0 | 0 | — |
case-19 | pass→fail | 8,379 | 7,524 | -10% | 1 | 1 | 0% | 1,356 | 604 | -55% | 0 | 0 | — |
case-20 | pass→pass | 15,863 | 14,278 | -10% | 1 | 1 | 0% | 2,958 | 3,220 | +9% | 0 | 0 | — |
case-21 | pass→pass | 8,813 | 3,821 | -57% | 1 | 1 | 0% | 1,511 | 937 | -38% | 0 | 0 | — |
case-22 | pass→pass | 14,510 | 9,902 | -32% | 1 | 1 | 0% | 2,554 | 2,084 | -18% | 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 8 counted toward the lift figure. The other 14 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 -5 percentage points is the difference between those two pass rates over the 8 comparable cases. 6 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/13/2026 | +59% |
Other measured skills in the registry, with their headline benchmark lift.