sensor-monitor: action-based triggers with capture/reference data passing

[Daniel Botros (danielbotros)]

What

Reworks sensor-monitor from "watch a sensor and notify via a fixed notifier" into a general action model: when a rule triggers or resolves, it fires DoCommands on any resources you name.

• Each rule has on_trigger and on_resolve action lists; each action is { "resource": <name>, "command": <DoCommand payload> } targeting any component or service.
• Notifying is just an action — point one at a notification service with { "command": "send", ... }. The dedicated notifier/message/sendNotification path is removed; the only fixed dependency is the watched sensor.
• on_trigger fires on the triggering edge and repeats per cooldown_seconds while triggered (default 0 = edge-only); on_resolve fires on the clearing edge.
• Action resources are declared as dependencies in Validate and resolved at construction by short name across any API (every resource.Resource implements DoCommand).

Capture & references (carrying data between commands)

Command string values may embed {{...}} references, resolved just before sending:

• Rule/reading context: {{value}}, {{key}}, {{threshold}}, {{operator}} (this also replaces the old {value} message templating).
• Captured responses: an action with "capture": "<name>" stores its DoCommand response; later actions reference its fields as {{name.field}}.

A value that is exactly one reference keeps the referenced value's type; an embedded reference is substituted as text. An unresolved reference skips that action. This is how a Slack message sent on trigger is reacted to on resolve — capture: "msg" on the send, then {{msg.ts}} / {{msg.channel}} in the resolve react command. Captures are overwritten on each cooldown re-fire (so resolve uses the latest) and cleared on resolve.

{{...}} (not ${...}) is used deliberately: the Viam config reader reserves ${...} for its own placeholder replacement and rejects unknown placeholders before the module ever sees the config.

Tests & docs

resolveValue unit tests (typed/embedded/nested/missing), trigger+resolve firing, cooldown repeat, capture-carries-across-trigger-resolve, capture-missing skip, and unresolved-dependency build failure. README rewritten for the action model.

Notes

• Breaking config change: the old notifier + per-rule message fields are gone. Existing configs migrate by moving each message into an on_trigger send action (with {{value}} in place of {value}).
• Depends on a notifier that supports a react command for the reaction use case — see Add "react" command for emoji reactions notifications#4. The bot token needs the reactions:write scope.
• Supersedes the closed react-on-resolve PR (sensor-monitor: react to alert message when a rule resolves #8).

🤖 Generated with Claude Code