Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Follow a structured recovery decision tree when tool calls fail instead of blindly retrying or giving up.
.claude/skills/aden-hive-hive-error-recovery/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -45% | 0% |
When a tool call fails:
Never silently drop a failed item. If the item is a task in the colony queue, write the failure to the DB instead of an in-memory buffer:
bashsqlite3 "$DB_PATH" "UPDATE tasks SET status='failed', last_error='<one-sentence reason>', completed_at=datetime('now'), updated_at=datetime('now') WHERE id='<task-id>' AND worker_id='<your-worker-id>';"
The tasks.retry_count column and the stale-claim reclaimer handle auto-retry for crashes; your job is the within-run decision tree above. See hive.colony-progress-tracker for the full queue protocol.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 13,873 | 6,331 | -54% | 1 | 1 | 0% | 2,800 | 1,467 | -48% | 0 | 0 | — |
case-02 | fail→pass | 11,218 | 4,956 | -56% | 1 | 1 | 0% | 2,037 | 1,196 | -41% | 0 | 0 | — |
case-03 | fail→pass | 9,984 | 4,168 | -58% | 1 | 1 | 0% | 1,870 | 1,056 | -44% | 0 | 0 | — |
case-04 | fail→pass | 10,719 | 4,737 | -56% | 1 | 1 | 0% | 1,864 | 1,246 | -33% | 0 | 0 | — |
case-05 | fail→fail | 7,083 | 3,566 | -50% | 1 | 1 | 0% | 1,230 | 1,013 | -18% | 0 | 0 | — |
case-06 | fail→pass | 12,403 | 4,339 | -65% | 1 | 1 | 0% | 2,087 | 1,154 | -45% | 0 | 0 | — |
case-07 | fail→pass | 10,332 | 4,755 | -54% | 1 | 1 | 0% | 1,671 | 1,178 | -30% | 0 | 0 | — |
case-08 | pass→pass | 5,864 | 4,478 | -24% | 1 | 1 | 0% | 954 | 1,041 | +9% | 0 | 0 | — |
case-09 | fail→pass | 4,917 | 3,116 | -37% | 1 | 1 | 0% | 961 | 990 | +3% | 0 | 0 | — |
case-10 | pass→fail | 10,108 | 4,899 | -52% | 1 | 1 | 0% | 1,677 | 1,052 | -37% | 0 | 0 | — |
case-11 | pass→pass | 11,963 | 5,408 | -55% | 1 | 1 | 0% | 1,968 | 1,211 | -38% | 0 | 0 | — |
case-12 | pass→pass | 9,163 | 3,552 | -61% | 1 | 1 | 0% | 1,597 | 940 | -41% | 0 | 0 | — |
case-13 | fail→pass | 8,657 | 2,578 | -70% | 1 | 1 | 0% | 1,518 | 721 | -53% | 0 | 0 | — |
case-14 | pass→pass | 7,365 | 4,122 | -44% | 1 | 1 | 0% | 1,310 | 1,089 | -17% | 0 | 0 | — |
case-15 | fail→pass | 5,378 | 4,213 | -22% | 1 | 1 | 0% | 869 | 1,079 | +24% | 0 | 0 | — |
case-16 | pass→pass | 3,720 | 3,563 | -4% | 1 | 1 | 0% | 611 | 903 | +48% | 0 | 0 | — |
case-17 | fail→fail | 9,879 | 1,582 | -84% | 1 | 1 | 0% | 1,821 | 583 | -68% | 0 | 0 | — |
case-18 | fail→pass | 11,371 | 1,420 | -88% | 1 | 1 | 0% | 2,151 | 560 | -74% | 0 | 0 | — |
case-19 | fail→pass | 9,706 | 5,640 | -42% | 1 | 1 | 0% | 1,726 | 1,385 | -20% | 0 | 0 | — |
case-20 | pass→pass | 13,615 | 9,793 | -28% | 1 | 1 | 0% | 2,537 | 2,196 | -13% | 0 | 0 | — |
case-21 | fail→pass | 7,544 | 7,720 | +2% | 1 | 1 | 0% | 1,442 | 1,781 | +24% | 0 | 0 | — |
case-22 | pass→pass | 10,076 | 7,198 | -29% | 1 | 1 | 0% | 1,934 | 1,562 | -19% | 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 +50 percentage points is the difference between those two pass rates over the 22 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.