swmm-plot — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited swmm-plot (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
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.
When the user asks to plot, always ask these questions first before calling any plot tool:
Link Flow Summary so the user can pick.Total_inflow, Depth_above_invert, Volume_stored_ponded, Flow_lost_flooding. Link options: Flow_rate, Velocity, Depth.Do not silently pick defaults. The user needs control: different plots answer different questions (peak inspection vs continuity vs flooding).
| 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.
| 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 |
apply_style(); size with figsize("single" | "double") or map_figsize(bounds); save with save_figure(fig, out_png). Never plt.rcParams.update, never tight_layout(), never bbox_inches="tight" (it silently changes the physical width and breaks the 89/183 mm rule).legend_outside), so it never sits on data.Rainfall depth (mm/5 min), Flow (m³/s), Easting (m, EPSG:32610), X (INP units) when the INP's unit is unknown.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.
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>.pdfCall 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 bboxThe 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).
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.pngThis 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).
.inp / .out path) and return the available rainfall series names, node IDs, and node output attributes. Call this before plot_run so you can pass real names instead of placeholders. Required args: run_dir (or inp_path + out_file). Read-only; auto-approved under the QUICK permission profile.inp to override. Required arg: run_dir. Optional: out_png, dpi, no_subcatchments, no_vertices.plot_rain_runoff_si on the MCP server): create the paired rainfall + flow figure from a run directory. Required arg: run_dir. Supply either node or link (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 where swmm-report looks for embeddable figures; a RELATIVE out_png is anchored there too (never the process working directory), while an absolute path is honored verbatim. Day-window cropping: pass focus_day (YYYY-MM-DD) to crop the axis to one calendar day; pass window_start and window_end (both HH:MM) to further narrow to a sub-day window; both require focus_day (the server rejects window_start/window_end without focus_day).`mcp/swmm-plot/server.js` exposes one underlying tool:
plot_run. Prefer plot_run (which accepts run_dir) over calling this directly.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 as outPdf.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 via inspect_plot_options): name of the rainfall TIMESERIES inside the .inp.rainKind (default "depth_mm_per_dt"): one of intensity_mm_per_hr, depth_mm_per_dt, cumulative_depth_mm.dtMin (default 5): 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 via inspect_plot_options): node ID to plot from the .out.nodeAttr (default "Total_inflow"): which swmmtoolbox attribute (e.g. Total_inflow, Lateral_inflow, Flow_lost_flooding).link (optional): conduit id; when set, the lower panel plots the link's Flow_rate instead of a node attribute. Mutually exclusive with node.width (default "single"): single (89 mm) or double (183 mm).dpi (default 450): 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 with focusDay): sub-day time window within the focus day. Rejected with a clear error if used without focusDay.padHours (default 2): padding around the rainfall extent when no focusDay is given.rainTs is a single name); multi-gauge inputs need separate figures.swmm-uncertainty / swmm-calibration, not here.~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.