swmm-modeling-memorylisted

# SWMM Modeling Memory ## What this skill provides - A downstream memory layer for audited Agentic SWMM runs. - Deterministic summaries of repeated assumptions, QA issues, failures, missing evidence, and run-to-run differences. - Run-level `memory_summary.json` cards that compress audit artifacts into reusable next-run context. - Project/case-level memory groups that keep Tod Creek, Tecnopolo, TUFLOW, Generate_SWMM_inp, acceptance, and other cases separate. - Summaries of deterministic SWMM-specific diagnostics when `model_diagnostics.json` is present. - Human-readable lessons learned from previous audit records. - Controlled skill update proposals that require human review and benchmark verification. This skill does not run SWMM, build SWMM models, modify existing skills, or claim autonomous self-improvement. Agentic SWMM is not only an automation workflow. It is a memory-informed, verification-first modeling system that can learn from audited modeling history through controlled skill refinement. ## When to use this skill Use this skill after `swmm-experiment-audit` has produced run-level artifacts such as: - `experiment_provenance.json` - `comparison.json` - `experiment_note.md` - `model_diagnostics.json` when available Use it when: - multiple audited runs exist, - the user wants lessons learned across runs, - the user asks for recurring failure patterns or QA issues, - the user wants evidence-informed skill refinement proposals. The proposals may point to relevan