oee-analysis - SKILL.md Agent Skill

name: oee-analysis description: Compute and plot OEE — a 7-day trend chart with target/threshold lines, an overlay of telemetry alarms and downtime stops, plus a flat CSV export. Saves /tmp/oee-trend--line--.png + .csv. enabled: true

Instructions

Use this skill whenever Frank asks for an OEE trend, a chart, "what does the week look like?", or any "show me, don't tell me" view of production. This is the computation surface — it uses code_interpreter against the numbers returned by azure-iot. Do NOT do mental math on OEE values; always send the numbers through this skill.

Workflow

Pull OEE rows via iot_get_oee(plant_id="P-CLE", line="<line>", date_from="<7 days ago>", date_to="<today>").
Pull downtime events via iot_list_downtime_events(plant_id="P-CLE", line="<line>", since="<7 days ago>") — used as tick marks on the chart.
(Optional, only for the spindle story) Pull telemetry via iot_get_telemetry("DEV-3001", since="<window>") — used as the secondary axis on the chart.
Pass the three JSON blobs into code_interpreter with the analysis script — DO NOT retype any numbers.
The script writes:
- /tmp/oee-trend-<plant_slug>-line-<N>-<date>.png — chart
- /tmp/oee-trend-<plant_slug>-line-<N>-<date>.csv — flat OEE for the brief
Reference both file paths in your response so file-sharing picks them up.

Reference script (oee_analysis.py)

The full script is in scripts/oee_analysis.py in this skill's directory. Read it once with file_read, then invoke via code_interpreter — DO NOT inline the script body in your response.

The script signature:

python /app/use-cases/plant-floor-supervisor/skills/oee-analysis/scripts/oee_analysis.py \
  --plant-id P-CLE \
  --line "Line 3 — Precision" \
  --date 2026-06-09 \
  --oee-json '<JSON returned by iot_get_oee>' \
  --downtime-json '<JSON returned by iot_list_downtime_events, optional>' \
  --telemetry-json '<JSON returned by iot_get_telemetry, optional>' \
  --out-dir /tmp

Outputs to stdout: a 1-line summary of trend direction and the file paths. Outputs to disk: the .png + .csv.

Chart conventions

Solid line: daily OEE % over the window
Dashed horizontal: target (e.g. 80%)
Shaded band: watch zone (target-5 to target), light yellow
Shaded band below: investigate zone (<target-5), light red
Vertical tick marks: each downtime event (height = duration_minutes / 10, max 5px)
Secondary axis (only if --telemetry-json): vibration overlay on the same x-axis — for the spindle story

When NOT to use

A single-day OEE number — just cite iot_get_oee inline; don't build the chart for one row.
Computing variance on financial spend (that's a different persona).