Requesting agent name

• main

North Star

Make GameDriver easy to use, flexible/powerful, and strongly debuggable/traceable.

Product goals

1. Simple authoring API for game scripts (low ceremony, clear defaults).
2. Composable engine primitives (resolver, scene classifier, recovery policy) for power/flexibility.
3. First-class observability (typed event stream + correlation IDs + deterministic reason enums).
4. Deterministic replay/debug loop with bounded telemetry and failure isolation.

Success criteria

• New strategy can be created with minimal boilerplate and pass contract tests.
• No duplicate click/recovery stacks in strategy code.
• Every decision/action has typed reason + correlation id in event stream.
• Operators can diagnose stuck/no-progress from logs/events without screen recording.

Execution tracks

• Track A: Usability API
  • Script-facing API simplification (clear TargetSpec/helpers)
  • Docs + examples
• Track B: Flexibility/Power
  • Canonical resolver + scene classifier + recovery state machine
• Track C: Debuggability/Traceability
  • Event stream schema v1, stable enums, bounded evidence, replay hooks

Immediate next slices

1. Define/lock script API surface v1 and usage guide.
2. Finalize typed event envelope adoption across runtime decision points.
3. Add trace correlation/reason coverage checks in tests.
4. Add incident replay utility (minimal) from events stream.

Agent operating model

• game-architect: design + sequencing + acceptance gates
• survivor-operator: runtime pain discovery + validation feedback
• pr-governor: quality gate, merge velocity, duplicate prevention

Definition of done for each slice

• issue linked, PR linked, tests included, rollback note, telemetry impact documented.