AGENT SKILL

reflow-machine-maintenance-guidance

Use handbook-grounded definitions/limits plus thermocouple/MES/defect data to answer reflow maintenance & compliance questions and produce required JSON artifacts.

License 许可
Apache-2.0
最近更新
2026-11-14
Stars
★ 0
Forks
1

01 前言 Front Matter

name
reflow-machine-maintenance-guidance
description
Use handbook-grounded definitions/limits plus thermocouple/MES/defect data to answer reflow maintenance & compliance questions and produce required JSON artifacts.

03 SKILL.md 正文

Steps

  1. Start by inspecting available inputs with plain ASCII shell flags/quotes (avoid copy/paste of malformed “-l”): list /app/data and preview CSV schemas (headers + first rows) to confirm required columns like run_id, tc_id, time_s, temp_c.
  2. Extract the exact definitions and numeric requirements from handbook.pdf before calculating anything: preheat definition + applicable temperature/zone/time region, ramp equation, ramp limit; TAL threshold/window/min/max; peak requirement and how to choose the representative thermocouple (e.g., “cold spot”/minimum peak guidance); conveyor geometry + dwell/length rules.
  3. Convert handbook content into a compact cfg object (regions, thresholds, limits) and use it consistently for every computation and output file.
  4. Compute per-run metrics from thermocouple traces deterministically:
    • always sort by run_id, tc_id, time_s;
    • ignore segments with dt <= 0;
    • use finite-difference slope for ramp in the handbook’s defined region;
    • use linear interpolation for time-above-threshold metrics (TAL-type).
  5. Reduce multiple thermocouples to run-level results using the handbook’s “representative” guidance; if the handbook is not explicit, use deterministic tie-breaks:
    • max-type metric ⇒ pick max value, then smallest tc_id;
    • min-type (“coldest”) metric ⇒ pick min value, then smallest tc_id.
  6. Write the required outputs to /app/output/q01.json … /app/output/q05.json, rounding floats to 2 decimals, sorting by IDs ascending, and emitting null where data/segments are missing (do not guess).
  7. Before finishing, verify that all /app/output/q0*.json files exist and contain valid JSON (no NaN/Inf).

Expected Result

Handbook-backed calculations are performed end-to-end and the five required JSON files are actually written under /app/output/, with deterministic sensor selection, correct rounding/sorting, and no hallucinated values.

04 同仓库更多 Skills