Human-readable cockpit that reads originator-telemetry.jsonl, groups by traceId, and prints a compact per-run timeline with elapsed time, heartbeat count, chars streamed, and status. Use when you need to see what a lane is doing without reading raw JSONL, or when invoking lane-telemetry-cockpit to inspect lane run summaries.
Installs into .claude/skills of the current project.
Are you the author of Organ Lane Telemetry Cockpit?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/nexuslinkproductions-organ-lane-telemetry-cockpit-yuri-os)
---
name: organ-lane-telemetry-cockpit
description: "Human-readable cockpit that reads originator-telemetry.jsonl, groups by traceId, and prints a compact per-run timeline with elapsed time, heartbeat count, chars streamed, and status. Use when you need to see what a lane is doing without reading raw JSONL, or when invoking lane-telemetry-cockpit to inspect lane run summaries."
triggers:
- "organ-lane-telemetry-cockpit"
- "how do I use lane-telemetry-cockpit"
- "lane-telemetry-cockpit usage"
- "lane-telemetry-cockpit guide"
- "Lane Telemetry Cockpit (human-readable view over YURI_LANE_TELEMETRY)"
generated: true
source_node: "lane-telemetry-cockpit"
source_file: "_SYSTEM/Scripts/lane-telemetry-cockpit.mjs"
scope: harness
invocation: workflow
---
<!-- GENERATED from the canonical graph node "lane-telemetry-cockpit" (mechanism.guide) by _SYSTEM/Scripts/yuri-guide-project.mjs.
DO NOT hand-edit — edit _SYSTEM/organ-guides.json, then run: node _SYSTEM/Scripts/yuri-guide-seed.mjs && node _SYSTEM/Scripts/yuri-guide-project.mjs -->
# Organ Guide — Lane Telemetry Cockpit (human-readable view over YURI_LANE_TELEMETRY)
**Module:** `_SYSTEM/Scripts/lane-telemetry-cockpit.mjs` · **Layer:** Skills & Orchestration · **Invocation:** both · **CLI:** `(path) [--json] [--limit N]`
**Purpose.** Human-readable cockpit over the Originator lane-telemetry stream. JSONL is the audit substrate, not the operator experience: this reads originator-telemetry.jsonl, groups by traceId, COLLAPSES the noisy stream/stderr/stdout chunk events into counts, and prints a compact per-run timeline (start → model call → verification → exit) with elapsed, heartbeat count, chars streamed, and status — so the operator sees what a lane is doing without reading raw JSON.
## Exports
- `readTelemetry(file = TELEMETRY, opts = {})`
- in: optional telemetry file path + opts
- out: parsed telemetry events (malformed lines tolerated)
- `summarizeByTrace(events)`
- in: telemetry events
- out: per-traceId summary collapsing chunk events into counts (elapsed, heartbeats, chars, status)
- `renderCockpit(traces, opts)`
- in: summarized traces + render opts
- out: the compact human-readable cockpit string
## Security boundary
Strictly READ-ONLY — reads the telemetry JSONL and renders it; never mutates telemetry. Input line-count is bounded (limit guard). No protected-path access.
## When to use
Inspecting what a dispatched lane run actually did (timeline, model call, verification, exit) without parsing raw originator-telemetry JSON; triaging a slow/failed lane run.
## Gotchas
- It collapses stream/stderr/stdout chunk events into COUNTS — individual chunk payloads are not shown (by design, to stay compact).
- Grouping is by traceId; a run without a stable traceId will not group cleanly.
- Read-only over a live-appended file — a snapshot reflects the moment it was read.
## Session Notes
### 2026-06-13
- session: 40m | peak ctx: 0% | compacts: 0
- tools: Bash×824, Read×163, Edit×17, StructuredOutput×16, Write×8, TodoWrite×4, ToolSearch×2, ScheduleWakeup×2, Workflow×1, mcp×1, AskUserQuestion×1
- corrections: none
- errors: none