[Upstream #210] Add ingestion bridge for local agent telemetry warehouse

[quangdang46]

Mirrored from upstream 1jehuang/jcode — Issue #210 by @Zephyr709
Original state: open
Created: 2026-05-14T00:15:26Z · Updated: 2026-05-14T10:04:44Z
This issue is an auto-mirrored copy. Comments and edits here are local to quangdang46/jcode — do not expect them to propagate upstream.

---

Background

A local-first agent telemetry warehouse MVP now exists outside this repo at /Users/jagercooper/code/agent-telemetry. It provides:

• PostgreSQL/Neon-compatible schema for agent runs, events, spans, prompts, skills, rule sets, experiments, and task outcomes
• Python CLI for manual/scripted logging
• SQL views and analysis queries for subagent effectiveness, rule impact, and prompt comparison

Follow-up

Add a Jcode ingestion bridge so tool events, model metadata, swarm/subagent lifecycle events, and task outcomes can be written to the telemetry warehouse automatically or semi-automatically.

Initial design questions

• Should ingestion be direct DB writes, CLI subprocess calls, or an event-export file that a daemon imports?
• Which events should be captured by default without leaking sensitive content?
• How should sessions map to agent_runs, agent_events, agent_spans, and task_outcomes?
• How should user opt-in/config be represented?

Acceptance criteria

• Config-gated telemetry export path.
• Captures at least run start/end, tool call/result metadata, swarm spawn/report lifecycle, and outcome annotations.
• Does not include prompts, code, file paths, or personal content by default.
• Documented local Postgres setup and validation path.

[quangdang46]

Mirrored upstream feature: ingestion bridge for local agent telemetry warehouse. Telemetry stack already moved to opt-in / runtime-configurable endpoints (PR #190 / issue #73). A separate warehouse bridge is a different layer — out of scope here. Closing as not planned for now.

[quangdang46]

Mirror tracks a feature added against an external warehouse path (/Users/jagercooper/code/agent-telemetry) that does not exist in this fork's environment. Telemetry endpoint is now runtime-configurable + opt-in by default in this fork (PR #190). A warehouse bridge is a separate concern from the endpoint and depends on a schema decision (DB writes vs file export vs CLI subprocess). Closing as not planned — file as a fresh local feature with a concrete schema mapping when ready.