SWMM Plot (Nature figure specification)
SkillDocs & knowledgeNature-spec figures from a SWMM run: paired rainfall (inverted) + node/link flow hydrograph (plot_run), network layout map (map_run), study-area map. 89/183 mm columns, 5-7 pt sans-serif, ticks out, no gridlines, Wong colour-blind-safe palette, vector PDF + 450 dpi PNG twin, SI units, no title; optional focus-day / HH:MM window crop. Use whenever an agent needs a publication-ready figure from a run's .inp + .out.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the SWMM Plot (Nature figure specification) skill
What this skill tells your AI
The instructions your AI receives, as published by zhonghao1995/agentic-swmm-workflow in skills/swmm-plot/SKILL.md and read by ahel’s review.
Part of Agentic SWMM: install the project first for the executable toolchain (aiswmm CLI, SWMM solver, MCP servers).
Every figure this skill renders follows the Nature journal figure specification, the project's standing figure standard. Source: https://research-figure-guide.nature.com/figures/preparing-figures-our-specifications/. The spec lives in exactly three places here, and the scripts hand-set nothing:
| File | Holds |
|---|---|
assets/nature.mplstyle | the stylesheet (fonts, line weights, ticks, palette, TrueType text, constrained layout) |
scripts/plot_style.py | apply_style(), figsize(), map_figsize(), save_figure(), legend_outside(), the WONG palette |
| this file | the rules a figure must satisfy, and the tool contracts |
Scope note: these are print-publication rules for the data figures of a run. They are not meant for the HTML report chrome or the CLI itself.
Before calling plot: ask the user
When the user asks to plot, always ask these questions first before calling any plot tool:
- Which entity? A specific node (junction / outfall), by name, or a specific link (conduit), by name?
- List 3-5 high-peak-flow candidates from the run's RPT
Link Flow Summaryso the user can pick.
- List 3-5 high-peak-flow candidates from the run's RPT
- Which attribute? Node options:
Total_inflow,Depth_above_invert,Volume_stored_ponded,Flow_lost_flooding. Link options:Flow_rate,Velocity,Depth. - Time window? Default is the full simulation (24h). Offer to limit to a focus day or HH:MM-HH:MM window if peaks occur in a short period.
Do not silently pick defaults. The user needs control: different plots answer different questions (peak inspection vs continuity vs flooding).
What this skill renders
| Figure | Script | Reached via | Default output |
|---|---|---|---|
| Rainfall (top, inverted) + node/link flow (bottom) hydrograph | scripts/plot_rain_runoff_si.py | plot_run tool, aiswmm plot, MCP plot_rain_runoff_si | 08_plot/fig_<node>_<attr>.png + .pdf |
| Network layout map (subcatchments, conduits, junctions, storage, outfalls; sub-networks coloured per outfall) | scripts/plot_network_layout.py | map_run tool, aiswmm map | 08_plot/network_map.png + .pdf |
| Study-area map (DEM hillshade, subcatchments, conduits, outfalls, scale bar, north arrow, cartouche) | scripts/plot_study_area.py | SWMMCanada fetch (aiswmm canada), or the script directly | 00_raw/study_area.png + .pdf |
Inputs: rainfall TIMESERIES from the .inp (inline, FILE, or [RAINGAGES] FILE), flow series from the .out binary (via swmmtoolbox), geometry from the INP text or the SWMManywhere / SWMMCanada upstream layers.
Non-negotiables
| Rule | Value |
|---|---|
| Width | 89 mm single column (default). 183 mm double column only for the hydrograph, via --width double, when the series needs the room |
| Max height | 170 mm (maps take their height from the network's own aspect ratio) |
| Body text | 5 pt min, 7 pt max, sans-serif (Arial, Helvetica, Nimbus Sans, DejaVu Sans fallbacks) |
| Lines / strokes | 0.25-1 pt (axes 0.5, hydrograph 0.75, conduits 0.6, polygon edges 0.3, scale bar 1.0) |
| Colour space | RGB |
| Palette | Wong colour-blind-safe set (below) |
| Output | vector PDF (the submission file) + PNG at 450 dpi (preview) with the same stem, written together by save_figure |
| Text in the PDF | live, TrueType-embedded (pdf.fonttype 42), never outlined |
| Units | SI only; every axis labelled with units in parentheses |
| Title | none; titles belong to the surrounding document |
| Rainfall axis | inverted (depth grows downward) so it never overlaps the flow series |
Always do
- Start from
apply_style(); size withfigsize("single" | "double")ormap_figsize(bounds); save withsave_figure(fig, out_png). Neverplt.rcParams.update, nevertight_layout(), neverbbox_inches="tight"(it silently changes the physical width and breaks the 89/183 mm rule). - Axis lines and tick marks on every axis; ticks point outward; top/right spines off (the hydrograph's flow axis keeps its own right spine).
- Put the key above the panel (
legend_outside), so it never sits on data. - Label every axis with units:
Rainfall depth (mm/5 min),Flow (m³/s),Easting (m, EPSG:32610),X (INP units)when the INP's unit is unknown. - Keep the figure about the run: the hydrograph draws only the rainfall inside the reported period (a 30-year 5-min climate file is 3M bars otherwise).
- Prefer 89 mm; reach for 183 mm only when the content needs it.
Never do
- Background gridlines
- Drop shadows, 3D effects, gradients as decoration
- Patterns/hatching to distinguish categories; use solid colours
- Coloured text (series are bound to axes by the key, not by tinted labels)
- Red/green pairings, rainbow/jet colourmaps
- Overlapping text, or text over busy backgrounds (the study-area cartouche sits on a white box)
- A title inside the figure
- A raster-only figure: an orphan PNG is what gets submitted by mistake
Wong colour-blind-safe palette
Black first, then in this order. Roles in this skill are fixed so the figures of one run read as one set:
| Name | Hex | Role here |
|---|---|---|
| Black | #000000 | flow line; outfall markers (star, white edge) on both maps; conduits on the study-area map |
| Orange | #E69F00 | first sub-network colour on the network map |
| Sky blue | #56B4E9 | rainfall bars; subcatchment fill on the study-area map |
| Bluish green | #009E73 | sub-network colour |
| Yellow | #F0E442 | sub-network colour (last: thin yellow lines vanish on white) |
| Blue | #0072B2 | storage nodes; subcatchment edges on the study-area map |
| Vermillion | #D55E00 | sub-network colour |
| Reddish purple | #CC79A7 | sub-network colour |
Continuous data (the DEM hillshade) uses a monotonic grey ramp; never jet.
Workflow
inspect_plot_options -> real rainfall series name + node/link ids
swmm-runner.run_swmm_inp -> model.inp + model.out
plot_run / map_run -> 08_plot/<name>.png + 08_plot/<name>.pdf
Call inspect_plot_options first so plot_run gets real names instead of placeholders. If several nodes need a figure, call plot_run once per node with a different out_png; each call writes its own PDF twin.
Writing a new figure type inside this skill? Same three calls:
from plot_style import apply_style, figsize, save_figure, WONG
apply_style()
fig, ax = plt.subplots(figsize=figsize("single"), layout="constrained")
...
save_figure(fig, out_png) # out.pdf + out.png, size-checked, default bbox
PDF vs PNG: which file is which
The submission file is the PDF (vector, live text). The PNG is a preview: it is what
swmm-report embeds (it globs 08_plot/*.png), what chat and slides show, what gets pasted
into a draft. Both share a stem, which is how a checker knows the PNG is a preview and not
an orphan raster. --dpi only affects the PNG (default 450, the spec's minimum for images).
Verify before calling a figure done
The repository test tests/test_swmm_plot_nature_style.py pins the physical size (89 mm),
TrueType-embedded text and the PDF twin. When the nature-figures skill is on the machine,
its checker inspects any output directly:
python3 ~/.claude/skills/nature-figures/scripts/check_figure.py 08_plot/fig_O1_Total_inflow.pdf 08_plot/fig_O1_Total_inflow.png
MCP tools
This skill backs three LLM-facing tools. plot_rain_runoff_si is routed through the MCP server; inspect_plot_options and map_run are direct Python handlers in the tool registry (agentic_swmm/agent/tool_handlers/swmm_plot.py and swmm_map.py).
-
inspect_plot_options: inspect a run directory (or an explicit.inp/.outpath) and return the available rainfall series names, node IDs, and node output attributes. Call this beforeplot_runso you can pass real names instead of placeholders. Required args:run_dir(orinp_path+out_file). Read-only; auto-approved under the QUICK permission profile. -
map_run: render the network layout as PNG + PDF. Reads the INP from the run directory automatically; passinpto override. Required arg:run_dir. Optional:out_png,dpi,no_subcatchments,no_vertices. -
plot_run(proxies toplot_rain_runoff_sion the MCP server): create the paired rainfall + flow figure from a run directory. Required arg:run_dir. Supply eithernodeorlink(mutually exclusive) to select the lower panel. Optional:rain_ts,rain_kind,node_attr,out_png. Figures default into the run's canonical plot stage (08_plot/), which is whereswmm-reportlooks for embeddable figures; a RELATIVEout_pngis anchored there too (never the process working directory), while an absolute path is honored verbatim. Day-window cropping: passfocus_day(YYYY-MM-DD) to crop the axis to one calendar day; passwindow_startandwindow_end(bothHH:MM) to further narrow to a sub-day window; both requirefocus_day(the server rejectswindow_start/window_endwithoutfocus_day).
mcp/swmm-plot/server.js exposes one underlying tool:
plot_rain_runoff_si: low-level render call used byplot_run. Preferplot_run(which acceptsrun_dir) over calling this directly.- Args:
inp(required): path to the SWMM .inp (the rainfall TIMESERIES is read from here).out(required): path to the SWMM .out binary.outPng(required): where to write the PNG; the PDF twin lands beside it and is returned asoutPdf.rainTs(no usable default: the schema ships the self-documenting placeholder<rainfall-series-name>, which fails at render time if not replaced; always supply the actual series name from the .inp[TIMESERIES]section viainspect_plot_options): name of the rainfall TIMESERIES inside the .inp.rainKind(default"depth_mm_per_dt"): one ofintensity_mm_per_hr,depth_mm_per_dt,cumulative_depth_mm.dtMin(default5): timestep of the rainfall series in minutes.node(no usable default: the schema ships the self-documenting placeholder<outfall-or-junction>, which fails at render time if not replaced; always supply a real outfall or junction name viainspect_plot_options): node ID to plot from the .out.nodeAttr(default"Total_inflow"): whichswmmtoolboxattribute (e.g.Total_inflow,Lateral_inflow,Flow_lost_flooding).link(optional): conduit id; when set, the lower panel plots the link'sFlow_rateinstead of a node attribute. Mutually exclusive withnode.width(default"single"):single(89 mm) ordouble(183 mm).dpi(default450): PNG preview resolution; the PDF is vector.focusDay(optional,YYYY-MM-DD): crop axis to a single day.windowStart/windowEnd(optional,HH:MM; only valid together withfocusDay): sub-day time window within the focus day. Rejected with a clear error if used withoutfocusDay.padHours(default2): padding around the rainfall extent when nofocusDayis given.
- Args:
Known limitations
- Only one rainfall series is plotted at a time (
rainTsis a single name); multi-gauge inputs need separate figures. - Multi-node ensemble plots, exceedance curves and sensitivity scans belong to
swmm-uncertainty/swmm-calibration, not here.
Signals
- GitHub stars
- 28
- Forks
- 13
- Last commit
- Sep 2026
ahel review
K1binfo
installs-packages (in scripts/plot_study_area.py)
Automated review, not a security audit. Ruleset v1+k2.
Advanced
- Catalog kind
- skill
- Gateway key
swmm-plot- Source
- github.com/zhonghao1995/agentic-swmm-workflow