🧠 Identity & Work List

The agent carries durable identity and a work list across wakes. These are maintained via maintenance (between wakes, not during) and can be read back via snapshots or reconstructed from events.

Identity

The identity document is free-form prose injected at the start of every wake. It records:

• Durable anchors: dubsar_id, House name, timezone defaults
• Operating posture: Adelaide-default time parsing, one-off timer discipline, transparent fallback/disclosure on failures
• Known limitations: snapshot continuity caveat, timer history sparseness, credential workflow sharp edges
• Current capabilities: image version, pre-installed tooling, resource model, lifecycle state

Work List

The work list is a durable projection of standing tasks and obligations. Key behaviors:

• Updated by maintenance between wakes — read the current version on each wake
• Use snapshots (list_snapshots.py --kind worklist) to compare state across wakes
• Fall back to events + message_sent content when snapshot content is unavailable
• Not a structured task database unless you encoded structure in text

Wake Summaries

Recent wake summaries are injected at the start of each wake, providing a narrative of what happened in prior wakes. This helps maintain continuity when snapshot retrieval is inconsistent.

Current Time Context

Each wake receives current-time context with UTC time, Bel-local time (Australia/Adelaide), timezone, and local date/day-of-week. This anchors all temporal reasoning.

Between-Wakes Maintenance

The agent does not update its own identity or work list during a wake. Only the maintenance process (between wakes) can modify these projections. The plan tool can append observations for future maintenance consideration.