AgentSkillsCN

apm-ds-exp

APM DS项目的假设驱动实验流程。适用于/apm-experiment。

SKILL.md
--- frontmatter
name: apm-ds-exp
description: Hypothesis-driven experiment workflow for APM DS projects. Use for /apm-experiment.
compatibility: opencode

What I do

  • Define experiment selection, planning, and reporting.
  • Enforce experiment hygiene and Memory Bank updates.

Experiment workflow

  1. Read memory-bank/ARCHITECTURE.md, memory-bank/TASK.md, memory-bank/STATE.md.
  2. Choose a hypothesis (or use user-provided).
  3. Plan hyperparameters and compute strategy.
  4. Create experiments/EXP-XXX_<desc>/ with:
    • main_exp.py
    • config.py
    • REPORT.md
  5. Test the pipeline quickly, do smoke-test; do not run full training unless the user asks.
  6. Record results and update Memory Bank.

Template

Use references/EXPERIMENT_REPORT_TMP.md for REPORT.md.

Required updates

  • memory-bank/TASK.md (mark hypothesis tested, update Active Experiment)
  • memory-bank/STATE.md (experiment history, best model tracker)
  • Log experiment runs and metrics per apm-logs.