Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Send an interrupt signal to a working team agent. Use when an agent is actively running and you need to send it a message without having to wait until it has processed all other messages.
.claude/skills/platformplatform-team-interrupt/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-04 | ✓→✗ | ▼ Worse | 73% | 0% |
| case-06 | ✓→✗ | ▼ Worse | 132% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | -34% | 0% |
| case-01 | ✗→✗ | ＝ Same ✗ | -56% | 0% |
SendMessage is for idle/hibernated agents only. If the agent is working, the message sits in the queue and is processed later against stale context - answers will be obsolete and queued instructions may already be wrong. To reach a working agent, use Interrupt + SendMessage. Never queue status checks, redirects, or corrections behind active work.
developer-cli exactly as written - do not expand to an absolute worktree path):bash dotnet run --project developer-cli -- claude-command send-interrupt-signal --team <team> --agent <agent>
#<id> interrupt ID.SendMessage to the same agent with the body prefixed by that ID: #<id> <your instruction>
The ID links the signal to the correct queued message; the active agent skips stale messages until it sees the matching ID.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 9,739 | 9,943 | +2% | 1 | 1 | 0% | 1,013 | 448 | -56% | 0 | 0 | — |
case-02 | fail→fail | 4,881 | 8,568 | +76% | 1 | 1 | 0% | 716 | 526 | -27% | 0 | 0 | — |
case-03 | fail→fail | 11,783 | 8,820 | -25% | 1 | 1 | 0% | 793 | 841 | +6% | 0 | 0 | — |
case-04 | pass→fail | 7,019 | 11,070 | +58% | 1 | 1 | 0% | 702 | 1,217 | +73% | 0 | 0 | — |
case-10 | fail→fail | 15,101 | 8,701 | -42% | 1 | 1 | 0% | 1,378 | 458 | -67% | 0 | 0 | — |
case-05 | pass→pass | 6,915 | 13,558 | +96% | 1 | 1 | 0% | 1,060 | 704 | -34% | 0 | 0 | — |
case-06 | pass→fail | 4,441 | 20,874 | +370% | 1 | 1 | 0% | 596 | 1,382 | +132% | 0 | 0 | — |
case-07 | fail→fail | 8,200 | 7,538 | -8% | 1 | 1 | 0% | 1,402 | 943 | -33% | 0 | 0 | — |
case-08 | fail→fail | 6,160 | 5,779 | -6% | 1 | 1 | 0% | 193 | 457 | +137% | 0 | 0 | — |
case-09 | fail→pass | 4,012 | 2,896 | -28% | 1 | 1 | 0% | 648 | 526 | -19% | 0 | 0 | — |
case-11 | fail→fail | 12,003 | 7,590 | -37% | 1 | 1 | 0% | 1,201 | 460 | -62% | 0 | 0 | — |
case-12 | fail→fail | 6,128 | 4,558 | -26% | 1 | 1 | 0% | 894 | 497 | -44% | 0 | 0 | — |
case-13 | fail→fail | 6,038 | 5,709 | -5% | 1 | 1 | 0% | 964 | 605 | -37% | 0 | 0 | — |
case-14 | fail→fail | 14,795 | 7,372 | -50% | 1 | 1 | 0% | 716 | 724 | +1% | 0 | 0 | — |
case-15 | fail→fail | 7,062 | 8,327 | +18% | 1 | 1 | 0% | 1,062 | 1,073 | +1% | 0 | 0 | — |
case-16 | fail→fail | 6,667 | 5,013 | -25% | 1 | 1 | 0% | 1,079 | 539 | -50% | 0 | 0 | — |
case-17 | fail→fail | 20,573 | 9,541 | -54% | 1 | 1 | 0% | 1,167 | 444 | -62% | 0 | 0 | — |
case-18 | fail→fail | 7,506 | 4,707 | -37% | 1 | 1 | 0% | 1,192 | 444 | -63% | 0 | 0 | — |
case-19 | fail→fail | 3,717 | 8,901 | +139% | 1 | 1 | 0% | 557 | 492 | -12% | 0 | 0 | — |
case-20 | fail→fail | 4,381 | 6,559 | +50% | 1 | 1 | 0% | 626 | 432 | -31% | 0 | 0 | — |
case-21 | fail→fail | 6,679 | 8,061 | +21% | 1 | 1 | 0% | 1,077 | 412 | -62% | 0 | 0 | — |
case-22 | fail→fail | 5,391 | 4,854 | -10% | 1 | 1 | 0% | 815 | 490 | -40% | 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 2 counted toward the lift figure. The other 20 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 2 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.