Telemetry Events

Event types

Type	Source	Fields
`llm_call`	Runner	model, promptTokens, completionTokens, latencyMs, costUsd
`llm_error`	Runner	model, errorType, retryCount
`tool_call`	Runner	toolName, durationMs, success
`agent_complete`	Runner	exitReason (cost_limit, max_iterations, halt, error, done)
`context_compaction`	Runner	tokensBefore, tokensAfter
`guard_scan`	Guard	side (write/read/snapshot), score, action, findingsCount
`agent_spawn`	Orchestrator	agentId, role, type, model
`agent_exit`	Orchestrator	agentId, exitReason, durationMs
`agent_kill`	Orchestrator	agentId, killReason
`protocol_start`	Orchestrator	protocolName, agentCount
`protocol_end`	Orchestrator	durationMs, totalTokens, totalCostUsd
`dance_call`	WebSocket	toolName, callerRole, latencyMs

JSONL format

Each line in the telemetry file is a JSON object:

{"type":"llm_call","timestamp":"2026-02-15T10:30:00.000Z","model":"claude-sonnet","promptTokens":500,"completionTokens":200,"latencyMs":1500,"costUsd":0.003}
{"type":"tool_call","timestamp":"2026-02-15T10:30:01.500Z","toolName":"read_file","durationMs":12,"success":true}
{"type":"guard_scan","timestamp":"2026-02-15T10:30:01.600Z","side":"write","score":5,"action":"PASS","findingsCount":0}

Storage location

~/.honeyb/projects/<slug>/telemetry/<date>.jsonl

One file per day, auto-rotated.

Aggregation

The reporter aggregates events into 5-minute summaries:

{
  "period": "2026-02-15T10:30:00Z/2026-02-15T10:35:00Z",
  "counts": { "llm_call": 42, "tool_call": 18, "guard_scan": 60 },
  "totalPromptTokens": 21000,
  "totalCompletionTokens": 8400,
  "totalCostUsd": 0.15,
  "avgLatencyMs": 1200,
  "errorRate": 0.02,
  "guardBlocks": 0,
  "guardWarns": 3
}

Aggregation runs for both local-only and cloud modes (needed for getSnapshot()).

Cloud sync

Opt-in via environment variables:

export TELEMETRY_ENDPOINT=https://carapace.honeyb.dev/v1/telemetry
export TELEMETRY_API_KEY=your-key

Cloud receives aggregated summaries only, never individual events or raw content. Sends every 5 minutes.

CLI commands

# Local overview (counts, totals, averages)
wgl telemetry overview --local

# Recent events
wgl telemetry events --type llm_call --limit 20

# Daily breakdown
wgl telemetry timeline

# Cloud overview (if configured)
wgl telemetry overview

Dashboard integration

The incubator dashboard Metrics tab receives live telemetry via WebSocket:

Push interval: every 10 seconds
Message type: metrics
Data source: reporter.getSnapshot()
Display: stat cards, event breakdown bars, numeric averages