AgentSkillsCN

weppcloud-agent

帮助 WEPPcloud 用户搭建并运行项目，理解与解读 WEPP 结果（径流、泥沙输送、洪峰流量、水量平衡），对比不同情景，并起草一份清晰的技术报告。适用于当用户分享 WEPPcloud 运行链接或运行 ID，并询问输出结果的含义、如何下载/整理结果、如何总结发现，或如何基于 WEPPcloud 结果撰写报告时使用。

SKILL.md
--- frontmatter
name: weppcloud-agent
description: Help WEPPcloud users set up and run projects, understand and interpret WEPP results (runoff, sediment delivery, peak flows, water balance), compare scenarios, and draft a clear technical report. Use when a user shares a WEPPcloud run link or run ID and asks what the outputs mean, how to download/organize results, how to summarize findings, or how to write a report based on WEPPcloud results.

Weppcloud Agent

Workflow (use this order)

0) Ask for the minimum inputs

  • •Run link(s): ask the user to paste the WEPPcloud URL(s) from their browser (preferred) or the run ID(s).
  • •Goal: what decision question the run supports (risk, treatment comparison, post-fire response, etc.).
  • •Scenario set: what differs between runs (climate period, land use/management, disturbance severity, treatments).
  • •Deliverable: quick explanation, a comparison table, or a full narrative report.

If any critical context is missing (area, time window, scenario differences), ask 1–2 targeted questions and proceed with a provisional interpretation.

1) Help them run the project (WEPPcloud web app)

Keep this user-facing and UI-agnostic:

  • •Confirm the project/scenario inputs align with the user’s goal (AOI, soils/landcover/management, climate source + period, any disturbance/treatment layers).
  • •Suggest a minimal scenario matrix when comparing (change one thing at a time; keep everything else fixed).
  • •When a run fails or stalls: collect the visible error, where it occurs (step/page), and any downloadable logs/diagnostics the UI exposes.

See references/user-workflows.md.

2) Pull results into a “working set”

Ask the user to download/export the key artifacts WEPPcloud provides (tables + maps) and point you to them (upload, paste snippets, or describe where they are in the UI).

Minimum recommended working set for interpretation:

  • •A small set of summary tables (outlet delivery, water balance, event extremes if relevant).
  • •A note of the modeled area (for unit normalization).
  • •If comparing scenarios: the same artifacts for each run.

See references/outputs-and-units.md.

3) Summarize + interpret (be explicit about units and basis)

When explaining results, always state:

  • •Where (outlet vs hillslope vs channel; whole watershed vs subareas).
  • •When (time window; annual average vs event-based).
  • •Units and area basis (mm vs volume; kg/ha vs tonnes).

Use references/comparison-checklist.md when comparing scenarios.

4) Draft the report

Use assets/report-outline.md as the default structure. Populate with:

  • •“What changed between scenarios”
  • •“What changed in outputs” (tables/figures)
  • •“So what” interpretation (implications, tradeoffs, uncertainties)

Always include:

  • •Exact run links/IDs
  • •Inputs summary (what the run represents)
  • •Key metrics with units and definitions

Return periods (peak flow)

If the user asks about short return periods (2/5/10-year) and wants event-date comparisons, request the event-by-event export (ebe_pw0.txt) for each scenario and run:

  • •scripts/return_period_compare.py --burned-ebe <path> --undisturbed-ebe <path> --start-year <yyyy>

This produces CSV tables (CTA + AM) plus an overlay histogram of peak discharge events.

Operator-only notes (do not assume available)

If (and only if) the user is an internal operator with filesystem or stack access, the following references describe backend-level resolution and report tooling:

  • •references/operator-wctl-workflows.md
  • •references/operator-run-directory-layout.md
  • •references/operator-wepppy-reports.md
  • •references/operator-weppcloudr-rendering.md

Conventions (keep responses actionable)

  • •Always state the run link/ID(s) you are interpreting.
  • •When comparing runs, normalize: units, time window (water years), and area basis (mm vs volume; kg/ha vs tonnes).
  • •Prefer exporting intermediate tables (CSV) alongside the narrative so results are reproducible.
  • •Only mention a filesystem run_dir when you actually have backend access.