Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Notify the Horizon workspace user about completed work, findings, or needed decisions. Only works when HORIZON env var is set.
.claude/skills/peters-horizon-notify/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 177% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -40% | 0% |
You are running inside Horizon, a GPU-accelerated terminal workspace. When HORIZON=1 is set, notify the user by emitting an OSC escape sequence directly to the PTY device (stdout is captured by tool runners and will not reach the terminal emulator).
Use this command (works on Linux and macOS):
printf '\033]0;HORIZON_NOTIFY:%s:%s\007' "<severity>" "<message>" > "/dev/$(ps -o tty= -p $PPID | tr -d ' ')"
Severities: info, done, attention Keep messages under 80 chars.
Use this when you:
Related OSC API for terminal titles:
printf '\033]0;HORIZON_TITLE:set:%s\007' "<title>" > "/dev/$(ps -o tty= -p $PPID | tr -d ' ')"
printf '\033]0;HORIZON_TITLE:clear\007' > "/dev/$(ps -o tty= -p $PPID | tr -d ' ')"
A set title stays on the panel titlebar until you clear it. Later OSC 0 titles from the TUI (cwd, spinner, status) do not overwrite it.
Keep titles under 80 chars and avoid newlines.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 9,376 | 8,616 | -8% | 1 | 1 | 0% | 1,434 | 529 | -63% | 0 | 0 | — |
case-02 | fail→fail | 8,884 | 13,523 | +52% | 1 | 1 | 0% | 1,301 | 572 | -56% | 0 | 0 | — |
case-03 | fail→fail | 29,221 | 26,761 | -8% | 1 | 1 | 0% | 1,845 | 703 | -62% | 0 | 0 | — |
case-04 | fail→pass | 8,450 | 6,276 | -26% | 1 | 1 | 0% | 500 | 1,385 | +177% | 0 | 0 | — |
case-05 | fail→pass | 10,000 | 5,308 | -47% | 1 | 1 | 0% | 1,541 | 1,193 | -23% | 0 | 0 | — |
case-06 | fail→pass | 16,458 | 6,451 | -61% | 1 | 1 | 0% | 1,401 | 1,247 | -11% | 0 | 0 | — |
case-07 | fail→fail | 27,570 | 25,540 | -7% | 1 | 1 | 0% | 1,199 | 1,097 | -9% | 0 | 0 | — |
case-08 | fail→pass | 11,568 | 12,647 | +9% | 1 | 1 | 0% | 1,592 | 1,299 | -18% | 0 | 0 | — |
case-09 | fail→pass | 18,575 | 6,471 | -65% | 1 | 1 | 0% | 1,719 | 1,030 | -40% | 0 | 0 | — |
case-10 | fail→pass | 11,097 | 4,040 | -64% | 1 | 1 | 0% | 1,745 | 890 | -49% | 0 | 0 | — |
case-11 | fail→pass | 21,202 | 27,070 | +28% | 1 | 1 | 0% | 2,917 | 2,005 | -31% | 0 | 0 | — |
case-12 | fail→pass | 12,819 | 4,147 | -68% | 1 | 1 | 0% | 1,923 | 748 | -61% | 0 | 0 | — |
case-13 | pass→pass | 12,754 | 6,651 | -48% | 1 | 1 | 0% | 1,716 | 1,166 | -32% | 0 | 0 | — |
case-14 | fail→fail | 14,318 | 29,124 | +103% | 1 | 1 | 0% | 2,489 | 939 | -62% | 0 | 0 | — |
case-15 | fail→pass | 11,627 | 8,648 | -26% | 1 | 1 | 0% | 2,009 | 1,646 | -18% | 0 | 0 | — |
case-16 | pass→pass | 9,649 | 3,900 | -60% | 1 | 1 | 0% | 1,446 | 784 | -46% | 0 | 0 | — |
case-17 | pass→pass | 16,735 | 15,487 | -7% | 1 | 1 | 0% | 2,481 | 2,702 | +9% | 0 | 0 | — |
case-18 | pass→pass | 11,398 | 10,229 | -10% | 1 | 1 | 0% | 727 | 1,200 | +65% | 0 | 0 | — |
case-19 | pass→pass | 3,234 | 3,954 | +22% | 1 | 1 | 0% | 371 | 861 | +132% | 0 | 0 | — |
case-20 | pass→pass | 4,817 | 4,352 | -10% | 1 | 1 | 0% | 661 | 925 | +40% | 0 | 0 | — |
case-21 | pass→pass | 2,879 | 6,204 | +115% | 1 | 1 | 0% | 319 | 1,084 | +240% | 0 | 0 | — |
case-22 | pass→pass | 15,331 | 5,172 | -66% | 1 | 1 | 0% | 1,634 | 989 | -39% | 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 17 counted toward the lift figure. The other 5 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 +41 percentage points is the difference between those two pass rates over the 17 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 9/8/2026 | +59% |
Other measured skills in the registry, with their headline benchmark lift.