What problem does it solve? Claude CLI's console exporter writes telemetry to stdout, but stdout cannot be redirected without breaking the TUI, so token usage data was silently lost. This Skill runs a lightweight local OTLP HTTP/JSON collector that receives telemetry and writes it to JSON Lines files, enabling accurate token and cost tracking for both pipeline runs and main-window sessions. ## Core Features & Use Cases - Dual-mode collection: Pipeline mode spawns a collector on a dynamic port per phase, while main-window mode runs a persistent collector on fixed port 49200 with SessionStart/SessionEnd hooks. - Daily rotation with SSoT: The collector self-rotates JSONL files at UTC+8 day boundaries and atomically publishes the active file path in main-otel-info.json, so aggregation never loses or double-counts records. - Idempotent aggregation: A SessionEnd hook reads JSONL files with per-file marker offsets and updates workflow_executions token and cost fields in the database. - Use Case: You run multi-phase story pipelines and want per-phase token costs in your DevConsole dashboard; the collector captures every api_request event and the watchdog aggregates them into the database automatically. ## Quick Start Ask the AI to set up OTel token tracking for the main window by starting the micro collector on port 49200 and verifying that telemetry is written to the daily JSONL log.
