feat(intelligence): OpenTelemetry token usage collection pipeline

[rwhite27]

Summary

• Replaces SSH polling for LLM token counts with a push-based OTel pipeline — no more CollectMachineTelemetryJob SSH usage calls when enabled
• Agent containers auto-instrument Anthropic/OpenAI SDK calls via @traceloop/instrumentation-anthropic; spans flow to an OTel Collector that aggregates gen_ai.usage.* attributes and forwards metrics to a new Laravel OTLP receiver
• Data lands in the existing agent_usage_snapshots table with source = '<provider>_otel', fully backwards-compatible with SSH-polled rows

Key components

File	Role
docker/otel/collector.yaml	OTLP gRPC receiver → spanmetrics connector → OTLP/HTTP exporter to Laravel
docker/otel/otel-init.js	Node.js OTel SDK bootstrap injected via NODE_OPTIONS=--require in container Dockerfiles
docker-compose.yml	New otel-collector service (ports 4317/4318)
config/otel.php	enabled, collector_endpoint, internal_token feature flags
BaseDockerComposeBuilderService	Injects OTEL_EXPORTER_OTLP_ENDPOINT + OTEL_SERVICE_NAME into agent containers when enabled
OtlpAdapterController	Thin REST bridge (validates X-Internal-Token) forwarding OTLP JSON to GraphQL
ingestAgentTelemetry mutation	Lighthouse mutation resolver → OtlpUsageIngestionService
OtlpUsageIngestionService	Parses OTLP metrics JSON, resolves deployment by ID or container name, upserts AgentUsageSnapshot
CollectDeploymentUsageCommand	Skip SSH poll when otel.enabled = true
Kernel.php	Enable CollectAgentTelemetryCommand cron (was commented out)

Enabling

Add to .env:

OTEL_ENABLED=true
OTEL_INTERNAL_TOKEN=<random-secret>   # same value in both .env and docker-compose.yml

Restart the stack — otel-collector will start alongside the existing services.

Containers (Dockerfile change needed)

Add this layer to each agent container Dockerfile (openclaw, hermes):

RUN npm install --save \
    @opentelemetry/sdk-node \
    @opentelemetry/exporter-trace-otlp-grpc \
    @opentelemetry/resources \
    @traceloop/instrumentation-anthropic

COPY docker/otel/otel-init.js /app/otel-init.js
ENV NODE_OPTIONS="--require /app/otel-init.js"

Test plan

• Start stack with OTEL_ENABLED=true — otel-collector container is healthy
• Make an LLM call through a running agent — row appears in agent_usage_snapshots with source = 'openclaw_otel' within 30s
• Set OTEL_ENABLED=false — SSH polling still creates rows with source = 'openclaw_docker'
• Confirm no double-counting: both sources active → two separate rows per deployment/date
• Debug mode: uncomment debug exporter in collector.yaml, check docker logs otel-collector shows received spans

🤖 Generated with Claude Code