@qmilab/lodestar-trace

@qmilab/lodestar-trace

The read side of the epistemic chain. Consumes an Lodestar event log and produces a markdown trust report that explains what an agent observed, what it came to believe, which beliefs informed which actions, what happened, and whether anything was revised.

CLI

lodestar report <session-id> [--project <id>] [--log-root <path>] [--out <file>]

Defaults the log root to ./.lodestar/events. Will scan project directories under the log root if --project is not supplied.

Examples:

# Render to stdout
lodestar report session-1779551238212

# Write to a file (suitable for pasting into a GitHub issue)
lodestar report session-1779551238212 --out trust.md

# Inspect a different log root
lodestar report session-1779551238212 --log-root .lodestar/events

The package also exposes the binary directly:

bunx lodestar-report <session-id>

Library

import {
  loadSessionEvents,
  projectChain,
  renderReport,
  defaultLogRoot,
} from "@qmilab/lodestar-trace"

const { events, project_id } = await loadSessionEvents({
  logRoot: defaultLogRoot(),
  session_id: "session-1779551238212",
})
const projection = projectChain(events, { session_id: "session-1779551238212" })
const markdown = renderReport(projection)

Why this is a separate package

The append-only event log is the source of truth in Lodestar. The trace package treats it as such: every fact it surfaces is grounded in a specific event. That separation is what lets lodestar report work even when the agent process has exited and only the log remains.

What it does not do

• Real-time tailing of an active session (planned for v0.2).
• Rendering to HTML or JSON (markdown only in v0).
• Exporting to LangSmith / Langfuse / Phoenix (that's @qmilab/lodestar-otel-exporter).
• Single-writer enforcement on the underlying event log (that's Batch 3).