Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Inter-agent communication via the msg CLI. Use this when you need to send messages to other agent sessions, read incoming messages, or coordinate with other agents in tmux panes.
.claude/skills/pchalasani-msg-inter-agent-communication/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -84% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -70% | 0% |
You can communicate with other coding agent sessions (Claude Code or Codex CLI) using the msg CLI tool.
Before sending or receiving messages, register yourself:
bashmsg register <your-name>
This auto-detects your tmux pane. You only need to do this once per session.
Send a message directly to another agent:
bashmsg send <agent-name> "Your message here"
Send to multiple agents:
bashmsg send agent1,agent2 "Message for both of you"
bashmsg reply <agent-name> "Your reply here"
Check your inbox:
bashmsg inbox
This shows all unread messages grouped by thread and marks them as read.
bashmsg list # List registered agents msg threads # List active threads msg status # Check system health
receiving agent's session.
recipient understands without re-reading the full thread.
them in the message text rather than pasting large blocks.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,976 | 4,149 | -48% | 1 | 1 | 0% | 732 | 484 | -34% | 0 | 0 | — |
case-02 | fail→fail | 16,987 | 3,714 | -78% | 1 | 1 | 0% | 1,694 | 469 | -72% | 0 | 0 | — |
case-03 | fail→fail | 6,490 | 6,156 | -5% | 1 | 1 | 0% | 1,242 | 491 | -60% | 0 | 0 | — |
case-04 | fail→pass | 4,473 | 2,117 | -53% | 1 | 1 | 0% | 728 | 560 | -23% | 0 | 0 | — |
case-05 | pass→pass | 5,305 | 1,697 | -68% | 1 | 1 | 0% | 1,006 | 654 | -35% | 0 | 0 | — |
case-06 | fail→pass | 7,291 | 1,428 | -80% | 1 | 1 | 0% | 1,350 | 544 | -60% | 0 | 0 | — |
case-07 | fail→pass | 16,137 | 1,056 | -93% | 1 | 1 | 0% | 2,761 | 434 | -84% | 0 | 0 | — |
case-08 | fail→pass | 6,908 | 1,442 | -79% | 1 | 1 | 0% | 1,039 | 435 | -58% | 0 | 0 | — |
case-09 | fail→pass | 9,079 | 2,870 | -68% | 1 | 1 | 0% | 1,417 | 422 | -70% | 0 | 0 | — |
case-10 | fail→pass | 7,516 | 954 | -87% | 1 | 1 | 0% | 1,221 | 410 | -66% | 0 | 0 | — |
case-11 | pass→pass | 7,847 | 2,411 | -69% | 1 | 1 | 0% | 1,370 | 722 | -47% | 0 | 0 | — |
case-12 | fail→pass | 7,177 | 2,707 | -62% | 1 | 1 | 0% | 1,358 | 723 | -47% | 0 | 0 | — |
case-13 | pass→pass | 6,896 | 1,391 | -80% | 1 | 1 | 0% | 1,119 | 541 | -52% | 0 | 0 | — |
case-14 | fail→fail | 4,879 | 3,504 | -28% | 1 | 1 | 0% | 867 | 466 | -46% | 0 | 0 | — |
case-15 | fail→fail | 3,058 | 3,697 | +21% | 1 | 1 | 0% | 553 | 509 | -8% | 0 | 0 | — |
case-16 | fail→pass | 10,520 | 1,341 | -87% | 1 | 1 | 0% | 1,809 | 502 | -72% | 0 | 0 | — |
case-17 | fail→fail | 5,093 | 3,142 | -38% | 1 | 1 | 0% | 919 | 414 | -55% | 0 | 0 | — |
case-18 | fail→fail | 4,532 | 4,338 | -4% | 1 | 1 | 0% | 787 | 526 | -33% | 0 | 0 | — |
case-19 | pass→pass | 3,793 | 2,675 | -29% | 1 | 1 | 0% | 596 | 694 | +16% | 0 | 0 | — |
case-20 | pass→fail | 1,877 | 3,850 | +105% | 1 | 1 | 0% | 283 | 413 | +46% | 0 | 0 | — |
case-21 | pass→pass | 8,783 | 6,775 | -23% | 1 | 1 | 0% | 1,813 | 1,612 | -11% | 0 | 0 | — |
case-22 | fail→fail | 4,136 | 3,267 | -21% | 1 | 1 | 0% | 717 | 418 | -42% | 0 | 0 | — |
case-23 | fail→pass | 7,855 | 2,403 | -69% | 1 | 1 | 0% | 1,409 | 687 | -51% | 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. 23 cases were attempted, and 14 counted toward the lift figure. The other 9 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 +35 percentage points is the difference between those two pass rates over the 14 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.