Skip to content

v0.6.3a1

Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 17 May 04:32
· 161 commits to main since this release
dad2d1c

Architectural enhancement pre-release on top of v0.6.2-alpha. pip install aiswmm continues to deliver v0.6.1 per PEP 440.

pip install aiswmm==0.6.3a1   # pin this pre-release
pip install --pre aiswmm      # allow any pre-release

What's new (one-line summary)

Keyword matching that was scattered across 6 files is now consolidated in one deep module; the 2163-LOC tool_registry.py is being split along skill-family lines; RAG memory retrieval is now agent-callable; and the portability claim from v0.6.2-alpha is fully delivered (the cases/ directory now ships with reference templates).

Architectural changes

  • New deep module agentic_swmm/agent/intent_classifier.py (#121). 372 LOC + 31 new tests. Adding a new intent touches one file, not six.
  • tool_handlers/ package (#128 partial). 3 of 10 family slices extracted (web, demo, swmm_memory). Remaining 7 queued.
  • 600 LOC of dead handlers removed from single_shot.py (#127). 797 → 144 LOC.

New agent capabilities

  • RAG memory retrieval is agent-callable (#124 Part A). retrieve_memory ToolSpec + memory-retrieval intent.
  • intent_map.json:mcp_enabled_skills now lists all 11 servers.

Portability completeness

  • cases/ ships with reference fixtures (#122). v0.6.2-alpha claimed portability via case_meta.yaml but shipped empty; now closed.
  • JSON config now scanned by AST regression guard (#122). The intent_map.json:21 "tecnopolo" entry that #118's guard missed is gone.

Documentation / hygiene

  • README "preload path" fixed in SKILL.md (#123).
  • README validation-snapshot anchors resolve (#129).
  • 14 /Users/zhonghao references cleaned from 5 public docs (#126).
  • Plot script defaults are self-documenting (#125).

Security note

PR #135 (closing #124) originally also exposed an LID interface tied to protected research IP (skills/swmm-lid-optimization/, gitignored). PR #137 reverted the LID-specific surface before this tag was minted, so the published v0.6.3a1 artifact does not carry the LID exposure. The RAG memory portion is kept intact.

Documentation

Full retrospective with before/after tables, PR map, migration guide, and reproducibility footprint: docs/releases/v0.6.3-alpha.md.

Reproducibility footprint

  • PyPI: aiswmm==0.6.3a1
  • Git tag: v0.6.3a1
  • Pinned model snapshot: gpt-5.5-2026-04-23 (unchanged since v0.6.1)
  • Docker base ref: AGENTIC_SWMM_REF=v0.6.3a1