Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Proactively extract critical values from tool results into working notes before automatic context pruning destroys them.
.claude/skills/aden-hive-hive-context-preservation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -56% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -48% | 0% |
You operate under a finite context window. Older tool results WILL be pruned. Extract what you need while it's still in context.
Save-as-you-go. After any tool call producing information you'll need later, immediately extract the key data into _working_notes or _preserved_data. Do not rely on referring back to old tool results — once they're pruned they're gone.
What to extract:
Handoffs between tasks happen through progress.db, not through shared-buffer handoff blobs. When you finish a task, any state the next worker needs goes into the task row itself (steps.evidence, tasks.last_error, sop_checklist.note) — see hive.colony-progress-tracker. Use _working_notes for things the DB schema doesn't cover.
You will receive an alert when context reaches {{warn_at_usage_ratio_pct}}% — preserve immediately.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→pass | 11,844 | 11,685 | -1% | 1 | 1 | 0% | 2,221 | 2,247 | +1% | 0 | 0 | — |
case-06 | fail→pass | 10,754 | 4,261 | -60% | 1 | 1 | 0% | 2,001 | 1,122 | -44% | 0 | 0 | — |
case-01 | fail→fail | 15,160 | 4,823 | -68% | 1 | 1 | 0% | 3,297 | 625 | -81% | 0 | 0 | — |
case-02 | fail→fail | 4,507 | 2,870 | -36% | 1 | 1 | 0% | 656 | 403 | -39% | 0 | 0 | — |
case-03 | pass→pass | 13,154 | 15,034 | +14% | 1 | 1 | 0% | 2,307 | 2,680 | +16% | 0 | 0 | — |
case-04 | pass→pass | 6,016 | 6,651 | +11% | 1 | 1 | 0% | 1,145 | 1,563 | +37% | 0 | 0 | — |
case-07 | fail→pass | 11,208 | 3,525 | -69% | 1 | 1 | 0% | 2,013 | 957 | -52% | 0 | 0 | — |
case-08 | fail→fail | 11,042 | 5,563 | -50% | 1 | 1 | 0% | 1,740 | 1,184 | -32% | 0 | 0 | — |
case-09 | fail→pass | 9,162 | 2,402 | -74% | 1 | 1 | 0% | 1,602 | 697 | -56% | 0 | 0 | — |
case-10 | fail→pass | 9,122 | 2,942 | -68% | 1 | 1 | 0% | 1,529 | 737 | -52% | 0 | 0 | — |
case-11 | fail→pass | 10,427 | 4,016 | -61% | 1 | 1 | 0% | 1,904 | 984 | -48% | 0 | 0 | — |
case-12 | fail→pass | 8,555 | 3,296 | -61% | 1 | 1 | 0% | 1,441 | 796 | -45% | 0 | 0 | — |
case-13 | fail→pass | 11,957 | 2,881 | -76% | 1 | 1 | 0% | 2,030 | 738 | -64% | 0 | 0 | — |
case-14 | fail→pass | 9,269 | 3,296 | -64% | 1 | 1 | 0% | 1,531 | 877 | -43% | 0 | 0 | — |
case-15 | fail→fail | 10,408 | 15,128 | +45% | 1 | 1 | 0% | 1,851 | 1,958 | +6% | 0 | 0 | — |
case-16 | fail→pass | 9,906 | 6,151 | -38% | 1 | 1 | 0% | 1,878 | 1,546 | -18% | 0 | 0 | — |
case-17 | fail→pass | 7,606 | 3,791 | -50% | 1 | 1 | 0% | 1,363 | 848 | -38% | 0 | 0 | — |
case-18 | fail→pass | 11,207 | 7,035 | -37% | 1 | 1 | 0% | 1,897 | 1,481 | -22% | 0 | 0 | — |
case-19 | pass→pass | 13,334 | 2,575 | -81% | 1 | 1 | 0% | 2,052 | 755 | -63% | 0 | 0 | — |
case-20 | fail→pass | 10,372 | 1,568 | -85% | 1 | 1 | 0% | 1,768 | 538 | -70% | 0 | 0 | — |
case-21 | pass→pass | 13,327 | 4,386 | -67% | 1 | 1 | 0% | 2,111 | 947 | -55% | 0 | 0 | — |
case-22 | fail→pass | 12,816 | 3,495 | -73% | 1 | 1 | 0% | 2,050 | 889 | -57% | 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 20 counted toward the lift figure. The other 2 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 +59 percentage points is the difference between those two pass rates over the 20 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.