Schedules & triggers

Chat is the trigger you see. Most of the value of an agent team, though, is in the work that starts without you: the morning brief, the weekly review, the campaign that posts on schedule, the nightly performance pull. WilsonOS treats those as first-class — with a ledger, a scope, a destination, and a watchdog — rather than as scripts running in the dark.

Three triggers, one path

flowchart TB
  C["Chat — you ask"] --> R
  S["Schedule — a registered automation fires"] --> R
  G["Signal — a connected channel reports something"] --> R
  R["A run record is written first"] --> W["The agent works under a scoped credential"]
  W --> D["Result delivered to a conversation you chose"]
  W --> L["Receipt: what ran, what it cost, what happened"]

Whatever starts a run, the same things are true: the run is recorded before anything executes (so a missing record is positive evidence that nothing ran), the work happens under a credential scoped to one brand and one run, spending waits for approval rules, and the outcome is written down.

Schedules

Every agent owns a registry of its scheduled automations — what fires, when, in which brand, and where the result goes. The registry is the canonical source; the system’s clock is generated from it, never hand-edited.

A few things you can rely on:

• Scope is declared, not inherited. Each entry names the brand it works in and, explicitly, any external access it needs. A scheduled run never inherits broad credentials from the machine it runs on.
• A destination is required. An automation that reports to chat names the conversation it reports to. Without one it does not post silently somewhere; it fails loudly and the failure is surfaced.
• Approval-gated. Agents may add automations for themselves, defaulting to off, and ask before enabling. Pausing is preferred to deleting.
• Time zones are handled. Entries are written in your time zone and converted for the clock.

Typical automations: a daily driver for a brand (what’s on today, what needs a decision), Mia’s daily and weekly reviews, a nightly memory consolidation, a nightly measurement pull, a weekly feedback distillation, a channel sync that pulls in new signals.

Signals from connected channels

Some work starts because the world moved: a batch of new performance data landed, a post went live, a listing changed. A channel engine — one per growth channel, sharing a common core — ingests those signals on a schedule, diagnoses what changed, and proposes actions. Its proposals arrive in a conversation like anything else; creating or tuning content from them goes through the usual review. See Channel engines.

The guardrails do not depend on you watching

An unattended run cannot be less careful than a conversation, so the same rules apply:

• Announce the capability. The receipt names the skill or flow that ran.
• Quote before spending. A paid step in an automation is either pre-approved in the entry at a stated scope or it stops and asks in the destination conversation.
• Brand rules apply. The run boots the brand’s essence like any session.
• Memory is written through the same writers. A run’s durable facts go through the sanctioned path, scoped; a scheduled run never writes into a person’s private memory.

What watches the watchers

• A watchdog compares each automation’s expected cadence with what actually fired and raises “expected run did not fire”.
• Failed or missed fires from the last two days appear in the agent’s boot bundle, so the agent itself knows and can tell you.
• Alerts land on an operator signals feed; in-app notification is best-effort, never the safety net.
• Silence is treated as a defect. A calm dashboard is only trusted when the ledger says the runs happened.

live The per-agent registry, run-record-first execution, scoped credentials, required destinations, and the watchdog are live. direction Operating-system-level isolation for scheduled runs (which chat conversations already have) is designed and scheduled.

Related

How agents runThe runtime a scheduled run executes in.

ConversationsWhere automations deliver their results.

Channel enginesSignal-driven work, one engine per growth channel.

FlowsWhat a scheduled automation usually runs.