AIToday
Large Language ModelsZenn AI/MLPublished: Nov 22, 2026, 16:00 JST

Shelly v7.0.0 adds confirm-then-schedule for AI agents

Send on LINE
Shelly v7.0.0 adds confirm-then-schedule for AI agents

3 Key Points

  1. What happened

    In a 2026-06-19 commit (b5af2be32), Shelly added a flow where a natural-language request is parsed by parseAgentNL into a preview card, and a scheduled agent is registered only after the user confirms.

  2. Why it matters

    This confirm-then-schedule design keeps agents that would never fire from being silently registered, since the parser only outputs allowed cron formats or null when confidence is absent.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

Shelly's implementation, detailed across commits from mid-to-late June 2026, reflects a deliberate approach to making natural-language scheduling safe and controllable. The parser, parseAgentNL, is designed to be deterministic for both Japanese and English, reducing ambiguity in what gets registered. By restricting schedule outputs to three allowed cron formats or null, the system avoids silently creating agents that would never fire. The confirmation card step adds a human check before any scheduled agent is created, and the action choices (draft, notify, webhook, cli) exclude publish, enforcing draft-only behavior at the type level. These choices suggest a focus on preventing unintended autonomous actions, which may be relevant to developers building similar features.

The reliance on Android's AlarmManager brings platform-specific constraints. Official documentation notes that alarms can wake the device, but Doze mode, introduced in Android 6.0, defers standard alarms when the device is idle. The setExactAndAllowWhileIdle API allows delivery during low-power idle, but frequency limits apply, with discrepancies between official sources. Additionally, exact alarms require permission checks from Android 13 onward. Shelly's design separates the scheduling operation from the completion of information collection and summarization, acknowledging that waking the device does not guarantee the full task will run.

The lifecycle controls, including the circuit breaker and global stop, address failure scenarios. The circuit breaker disables agents after three consecutive failures, while skipped and cli-declined runs do not count toward that threshold. The global stop persists across reboots, restoring before startup repair. These mechanisms are presented as implemented in commits, though some device-level testing, such as Telegram real-reception, was deferred, indicating that the build-ready state was distinguished from full end-to-end verification.

FAQ
What happens if a scheduled agent fails repeatedly?
A circuit breaker automatically disables an agent after three consecutive failures, cancels its schedule, persists enabled=false, and sends a notification.
How does the global stop work?
haltAllAgents and resumeAllAgents cancel or re-register all schedules and combine with a .halted marker that persists across reboots.
What are the Android limitations for exact alarms?
Starting with Android 13 (API level 33), SCHEDULE_EXACT_ALARM is not pre-granted for new installs, and repeating alarms are inexact since Android 4.4 (API level 19).
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleASIO's Mike Burgess flags AI data centre foreign-interference risk