v0.9.0

What's Changed

• Tool routing, API reference, wiring recipes, description enrichment by @brianlball in #39
• Title: Harden test suite + fix Choice-type measure arg validation by @brianlball in #41
• Merge optimize → develop by @brianlball in #46

Full Changelog: 0.8.2...v0.9.0

v0.8.2

What's Changed

• Measure authoring improvements: ReportingMeasure, Choice args, descriptions, LLM guidance by @brianlball in #37

Full Changelog: 0.8.1...0.8.2

v0.8.1

What's Changed

• AI-assisted measure authoring + CooledBeam terminals + zone equipment priority by @brianlball in #34
• Measure authoring reliability + OS App compatibility + E2E test by @brianlball in #36

Full Changelog: 0.7.0...0.8.1

v0.7.0

What's Changed

• Optimize tool discovery: slim responses, generic access, remove 11 redundant tools (142→131) by @brianlball in #33

Full Changelog: 0.6.0...0.7.0

v0.6.0

What's Changed

• anti-loop guardrails by @brianlball in #26

Full Changelog: 0.5.0...0.6.0

v0.5.0

New Tools (3)

• create_bar_building — create rectangular building geometry from DOE prototype type, stories, and floor area using ComStock measures
• create_new_building — one-call building creation: chains empty model → bar geometry → weather/DDY → typical constructions/loads/HVAC
• import_floorspacejs — import custom geometry from FloorspaceJS JSON files with auto zone creation and surface matching

Breaking Changes

• set_weather_file removed — use change_building_location instead (sets EPW + DDY + climate zone in one call)

Bug Fixes

• Fix JSON-string list params on 9 tools — MCP clients sending "["x"]" instead of ["x"] now handled
• Fix EUI units — report MJ/m² + kBtu/ft² alongside GJ/m² for practical use
• Fix conditioned_floor_area — compute from model geometry instead of relying on SQL (which returned 0 without HVAC)
• Fix get_run_artifacts for measure-only runs (no simulation output)
• Fix run_qaqc_checks error handling when no simulation results exist
• Add report_path and user_message to viewer/artifact responses

Agent Reliability

• Agent guardrails — MCP-first instructions, skill deduplication, troubleshoot skill for debugging failed simulations
• Eval scenarios — "should trigger" / "should not trigger" tables for 8 skills
• Known-gotcha hints — tool descriptions now warn about common pitfalls (baseline system zone requirements, EPW needing
  .stat/.ddy companions, simulation prerequisites)
• change_building_location — added EPW directory path hint so agents can discover available weather files

LLM Test Suite (new, local-only)

• 90 tests across 6 tiers: setup, tool selection, workflows, eval cases, guardrails, progressive prompts
• Runs claude -p against Docker MCP server, parses NDJSON for tool call assertions
• Benchmark tracking: tokens, cost, timing per test with history
• Stable/flaky markers (pytest -m stable for CI-safe subset)
• 95%+ pass rate on sonnet with retries

Documentation

• docs/testing.md — framework guide with annotated examples and CI details
• Geometry workflow docs and skill catalog updates
• Archived stale planning docs

Full Changelog: 0.4.0...0.5.0

0.4.0

What's Changed

• Fix reporting measures + bundle flags + climate zone estimation by @brianlball in #12
• Fix reporting + HVAC chilled beam terminals by @brianlball in #14
• fix SWIG stdout leak in hvac_systems breaking MCP JSON-RPC by @brianlball in #16
• Context reduction: brief mode, auto-load, batch ops, description compression by @brianlball in #18

Full Changelog: 0.3.0...0.4.0

v0.3.0

What's Changed

• fix SWIG memory leak warning on MCP subprocess exit by @brianlball in #2
• fix swig mem leak in pytest by @brianlball in #3
• Add Claude Code skills, skill discovery tools, test cleanup by @brianlball in #5
• add @mcp.prompts() and resources by @brianlball in #6
• Hardening, lint, input validation, and HVAC supply auto-wiring by @brianlball in #8
• 0.3.0 by @brianlball in #9

Full Changelog: https://github.com/NatLabRockies/openstudio-mcp/commits/0.3.0