Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Zoom meeting control for Sutando's voice and phone agents — contributes the summon (join + screen share), dismiss (leave meeting), and join_zoom (join with computer audio, no share) inline tools.
.claude/skills/sonichi-zoom/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -45% | 0% |
Manifest-loaded skill that contributes three Zoom inline tools into the agent's runtime tool table. Picked up at startup by loadSkillManifestTools() in src/inline-tools.ts and merged into inlineTools + ownerOnlyTools, so the tools reach both the web voice agent and the phone agent for owner callers.
summon — opens Zoom (via the zoommtg:// deeplink), joins the meeting, retries the Join click, and starts screen sharing so the user can see and control the Mac remotely. Optionally dials in via the phone server for voice. Use for "summon", "share my screen", "start zoom", "let me see your screen".dismiss — leaves the current Zoom meeting (stops screen share, opens the leave dialog, confirms; force-kills Zoom if windows linger). Use for "dismiss", "leave zoom", "end meeting", "hang up zoom".join_zoom — joins a Zoom meeting via the desktop app with computer audio, no screen share. Use for "join the zoom", "join meeting", or when the user provides a meeting ID.ZOOM_DEFAULT_SHARE_SCREEN (non-secret, in manifest.json config) — "true" by default. Set to "false" in the host environment to make summon skip screen sharing unless the caller explicitly passes shareScreen: true.ZOOM_PERSONAL_MEETING_ID, ZOOM_PERSONAL_PASSCODE (secrets — host .env only, never in the manifest) — the personal-room defaults used when summon / join_zoom are called without an explicit meeting ID.src/meeting-tools.ts in issue #786.skills/zoom/tools.ts and the skill is genuinely optional.dismiss with its own implementation (SIGTERM self instead of Zoom AppleScript); its dedupe-by-name loop keeps the override and drops this skill's dismiss.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-08 | fail→pass | 9,010 | 1,988 | -78% | 1 | 1 | 0% | 1,372 | 833 | -39% | 0 | 0 | — |
case-18 | fail→pass | 8,122 | 2,355 | -71% | 1 | 1 | 0% | 1,239 | 824 | -33% | 0 | 0 | — |
case-01 | fail→fail | 2,666 | 6,367 | +139% | 1 | 1 | 0% | 325 | 736 | +126% | 0 | 0 | — |
case-02 | fail→fail | 2,848 | 5,079 | +78% | 1 | 1 | 0% | 339 | 734 | +117% | 0 | 0 | — |
case-03 | fail→fail | 2,708 | 3,298 | +22% | 1 | 1 | 0% | 357 | 654 | +83% | 0 | 0 | — |
case-04 | pass→pass | 10,327 | 1,614 | -84% | 1 | 1 | 0% | 1,650 | 709 | -57% | 0 | 0 | — |
case-05 | fail→pass | 12,444 | 1,865 | -85% | 1 | 1 | 0% | 1,719 | 813 | -53% | 0 | 0 | — |
case-06 | pass→pass | 7,151 | 2,116 | -70% | 1 | 1 | 0% | 992 | 833 | -16% | 0 | 0 | — |
case-07 | fail→pass | 11,856 | 2,955 | -75% | 1 | 1 | 0% | 1,766 | 830 | -53% | 0 | 0 | — |
case-09 | pass→pass | 10,325 | 2,015 | -80% | 1 | 1 | 0% | 1,448 | 755 | -48% | 0 | 0 | — |
case-10 | fail→pass | 12,969 | 3,550 | -73% | 1 | 1 | 0% | 1,936 | 1,068 | -45% | 0 | 0 | — |
case-11 | fail→pass | 14,438 | 2,557 | -82% | 1 | 1 | 0% | 2,051 | 888 | -57% | 0 | 0 | — |
case-12 | fail→pass | 6,145 | 1,641 | -73% | 1 | 1 | 0% | 999 | 753 | -25% | 0 | 0 | — |
case-19 | fail→pass | 3,350 | 1,458 | -56% | 1 | 1 | 0% | 485 | 724 | +49% | 0 | 0 | — |
case-13 | fail→pass | 23,235 | 1,533 | -93% | 1 | 1 | 0% | 3,647 | 751 | -79% | 0 | 0 | — |
case-14 | fail→pass | 7,785 | 1,746 | -78% | 1 | 1 | 0% | 1,178 | 775 | -34% | 0 | 0 | — |
case-15 | fail→fail | 11,713 | 1,656 | -86% | 1 | 1 | 0% | 1,992 | 755 | -62% | 0 | 0 | — |
case-16 | fail→pass | 25,212 | 1,351 | -95% | 1 | 1 | 0% | 4,108 | 708 | -83% | 0 | 0 | — |
case-17 | fail→pass | 15,269 | 2,365 | -85% | 1 | 1 | 0% | 2,229 | 838 | -62% | 0 | 0 | — |
case-20 | pass→pass | 3,771 | 3,586 | -5% | 1 | 1 | 0% | 567 | 994 | +75% | 0 | 0 | — |
case-21 | fail→pass | 8,744 | 3,343 | -62% | 1 | 1 | 0% | 1,288 | 935 | -27% | 0 | 0 | — |
case-22 | fail→pass | 6,631 | 3,426 | -48% | 1 | 1 | 0% | 1,057 | 1,038 | -2% | 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 19 counted toward the lift figure. The other 3 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 +64 percentage points is the difference between those two pass rates over the 19 comparable cases.
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.