Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Start the Agentlas Connect flow for Telegram.
.claude/skills/agentlas-ai-hep-connect/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -57% | 0% |
Update fallback: 자동 업데이트가 안 되면 hephaestus update를 한 번 실행하세요. 업데이트하지 않아도 현재 버전 명령은 그대로 동작합니다.
Raw arguments: everything the user typed after /skill:hep-connect.
Use this prompt when the operator wants Telegram to talk to one Agentlas target: a single agent, a saved group, a local team, or a .agentlas org chart.
Current product contract:
runtime-owned secret store.
/connect; it lists savedgroups, .agentlas org charts, local teams, team agents, and single agents.
group bindings use one session per chat. Single-agent direct chats can keep one session per user.
If running inside Agentlas Desktop development, report that the Desktop route is /connect and use the real local Connect UI as the source of truth. Do not pretend that Telegram delivery is live until token validation, chat pairing, and a test message pass.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 12,731 | 9,590 | -25% | 1 | 1 | 0% | 2,022 | 1,850 | -9% | 0 | 0 | — |
case-02 | fail→pass | 15,508 | 10,190 | -34% | 1 | 1 | 0% | 2,182 | 1,810 | -17% | 0 | 0 | — |
case-03 | fail→pass | 15,303 | 8,651 | -43% | 1 | 1 | 0% | 1,394 | 1,612 | +16% | 0 | 0 | — |
case-04 | pass→pass | 6,969 | 8,940 | +28% | 1 | 1 | 0% | 1,046 | 1,230 | +18% | 0 | 0 | — |
case-05 | fail→pass | 17,971 | 10,184 | -43% | 1 | 1 | 0% | 2,686 | 1,812 | -33% | 0 | 0 | — |
case-06 | pass→pass | 9,755 | 6,907 | -29% | 1 | 1 | 0% | 1,266 | 1,131 | -11% | 0 | 0 | — |
case-07 | fail→pass | 10,849 | 2,938 | -73% | 1 | 1 | 0% | 1,563 | 670 | -57% | 0 | 0 | — |
case-08 | fail→pass | 19,711 | 9,715 | -51% | 1 | 1 | 0% | 2,889 | 1,835 | -36% | 0 | 0 | — |
case-09 | fail→fail | 11,135 | 2,405 | -78% | 1 | 1 | 0% | 1,552 | 608 | -61% | 0 | 0 | — |
case-10 | fail→pass | 12,395 | 3,884 | -69% | 1 | 1 | 0% | 1,554 | 854 | -45% | 0 | 0 | — |
case-11 | fail→pass | 15,081 | 5,289 | -65% | 1 | 1 | 0% | 2,053 | 894 | -56% | 0 | 0 | — |
case-12 | fail→pass | 6,773 | 5,446 | -20% | 1 | 1 | 0% | 706 | 944 | +34% | 0 | 0 | — |
case-13 | fail→pass | 9,257 | 3,184 | -66% | 1 | 1 | 0% | 1,175 | 717 | -39% | 0 | 0 | — |
case-14 | fail→pass | 15,360 | 2,866 | -81% | 1 | 1 | 0% | 2,104 | 754 | -64% | 0 | 0 | — |
case-15 | pass→pass | 12,534 | 4,817 | -62% | 1 | 1 | 0% | 1,746 | 904 | -48% | 0 | 0 | — |
case-16 | pass→pass | 13,259 | 4,873 | -63% | 1 | 1 | 0% | 1,791 | 964 | -46% | 0 | 0 | — |
case-17 | fail→pass | 27,225 | 4,596 | -83% | 1 | 1 | 0% | 2,538 | 973 | -62% | 0 | 0 | — |
case-18 | pass→pass | 10,902 | 3,775 | -65% | 1 | 1 | 0% | 1,477 | 957 | -35% | 0 | 0 | — |
case-19 | pass→pass | 27,564 | 17,673 | -36% | 1 | 1 | 0% | 2,074 | 1,398 | -33% | 0 | 0 | — |
case-20 | pass→pass | 12,892 | 3,206 | -75% | 1 | 1 | 0% | 1,950 | 694 | -64% | 0 | 0 | — |
case-21 | pass→pass | 14,207 | 6,887 | -52% | 1 | 1 | 0% | 2,034 | 1,135 | -44% | 0 | 0 | — |
case-22 | pass→pass | 12,830 | 3,829 | -70% | 1 | 1 | 0% | 1,652 | 807 | -51% | 0 | 0 | — |
case-23 | pass→fail | 13,321 | 8,296 | -38% | 1 | 1 | 0% | 1,764 | 1,568 | -11% | 0 | 0 | — |
case-24 | fail→fail | 15,169 | 16,592 | +9% | 1 | 1 | 0% | 2,476 | 2,848 | +15% | 0 | 0 | — |
case-25 | pass→pass | 15,161 | 11,494 | -24% | 1 | 1 | 0% | 2,395 | 2,272 | -5% | 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. 25 cases were attempted. The headline lift of +44 percentage points is the difference between those two pass rates over the 25 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.