Agent session replay: store last N events per agent for debugging

[leocagli]

Context

When an agent goes offline or fails a quest, there's no way to see what events led to the failure. A per-agent event replay buffer (last 50 events) enables post-mortem debugging without external logging infrastructure.

What to implement

Ring buffer per agent

Store the last 50 OS events (any type) for each agentId:

interface ReplayEvent {
  type: string        // e.g. 'agent.status', 'quest.completed'
  payload: object
  recordedAt: string  // ISO timestamp
}

Implemented as a fixed-size ring buffer — oldest entry dropped when full.

Route: GET /api/agents/:id/replay

Returns the agent's event buffer, newest first:

{ "agentId": "bot-1", "events": [...], "bufferSize": 50, "count": 23 }

Route: DELETE /api/agents/:id/replay

Clears the buffer (useful for test isolation).

Acceptance criteria

• Events are recorded as they fire through the OS event bus
• Buffer wraps at 50 entries — 51st event evicts the oldest
• GET returns events newest-first
• DELETE clears the buffer; subsequent GET returns count: 0
• Agent with no events returns { events: [], count: 0 }
• Unit tests: fill buffer past 50, wrap behaviour, clear

🎁 Evidencia visual = reward extra

• Si tu PR incluye video demo o capturas de pantalla mostrando la funcionalidad pedida funcionando end-to-end, va a ser considerado para rewards de GrantFox en esta issue. No es obligatorio, pero suma mucho para la evaluación.

[aliyuHabibu]

Hi, Please assign the issue to me