agent-orchestrator-mcp-server 0.7.10 → 0.7.12
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json
CHANGED
|
@@ -91,6 +91,19 @@ This tool creates a one-time AO wake-up trigger bound to the target session. The
|
|
|
91
91
|
- Use IANA timezone names (e.g., "America/New_York", "Europe/London", "Asia/Tokyo"). Do NOT pass UTC offsets like "+05:00" in the timezone parameter.
|
|
92
92
|
- If you omit timezone, wake_at is treated as UTC.
|
|
93
93
|
|
|
94
|
+
**Choosing wake_at — adaptive scheduling for unknown durations:**
|
|
95
|
+
When monitoring downstream work whose duration you can't predict (e.g., a subagent or pipeline phase), the bias is **prefer over-polling to under-polling**. A too-frequent poll wastes a few seconds of compute; a too-long sleep wastes minutes of user-facing wall-clock time and erodes trust. When in doubt, go shorter.
|
|
96
|
+
|
|
97
|
+
**Rules:**
|
|
98
|
+
- **First wake: MUST be ≤5 minutes from now.** Use less if you have any reason to think the work could already be done (e.g., a 30-second task — pick 1–2 minutes). This is a hard cap, not a default. Do NOT pick a longer first wake just because the work "might take a while" — you have not observed anything yet, so you cannot know.
|
|
99
|
+
- **Second and later wakes:** Now that you've actually observed progress, you may scale the next wake to what you saw:
|
|
100
|
+
- If the work is nearly done (~80%+): a few more minutes.
|
|
101
|
+
- In between: proportional to remaining work, capped at ~15 minutes.
|
|
102
|
+
- If barely started (<20%) AND you have already polled at least twice and confirmed the work is genuinely long-running: you may extend up to ~30 minutes. Do NOT use this tier on the first or second poll.
|
|
103
|
+
- **Never** pick a wake interval ≥10× the expected total task duration. If a downstream task should take ~3 minutes, a 25-minute wake is wrong even on the first poll — pick 2–3 minutes instead.
|
|
104
|
+
|
|
105
|
+
This guidance does NOT apply when waking at a known wall-clock time (e.g., "9am tomorrow") — use the calculated time directly.
|
|
106
|
+
|
|
94
107
|
**Parameters:**
|
|
95
108
|
- **session_id**: The session to wake up. Works from either \`needs_input\` or \`running\` state — if you call this tool from within your own currently-running session, the sleep transition is recorded and takes effect after the current turn ends.
|
|
96
109
|
- **wake_at**: ISO 8601 datetime without offset for when to wake up (e.g., "2026-04-15T14:30:00")
|