- Add server-side integer coercion for numeric parameters in all tools
- Fix parameter type validation issues (read_resource, find_in_file, read_console, manage_scene, manage_asset)
- Add proper tool descriptions with ctx parameter documentation
- Fix Context type annotations (use Context instead of Any for ctx)
- All tools now accept flexible numeric inputs (strings, floats) and coerce to integers
- Telemetry system working with all tool_execution events captured in BigQuery
- Remove invalid parameter type warnings from client-side validation
* commit '3e83f993bfe632034bf7302d4319e3cd16353eb8':
Improved ci prompt testing suite (#270)
chore: bump version to 3.3.2
Fix: Unity Editor reload crash + debug-noise reduction (#266)
Revise README for improved clarity and organization
docs: install uv via official installer (curl/winget)
Update README.md
docs: fix Windows uv path to use WinGet shim, keep macOS AppSupport symlink path
docs: update README.md with improved installation paths, documentation, and logo
fix: Update README installation paths to match ServerInstaller.cs
* CI: streamline Unity licensing (ULF/EBL); drop cache mounts & EBL-in-container; NL suite: clarify T-E/T-J, anchor positions, EOF brace targeting, SHA preconditions
* CI: support both ULF + EBL; validate ULF before -manualLicenseFile; robust readiness wait; use game-ci @v2 actions
* CI: activate EBL via container using UNITY_IMAGE; fix readiness regex grouping
* CI: minimal patch — guard manualLicenseFile by ulf.ok, expand error patterns, keep return-license @v2 for linter
* CI: harden ULF staging (printf+chmod); pass ULF_OK via env; use manual_args array for -manualLicenseFile
* CI: assert EBL activation writes entitlement to host mount; fail fast if missing
* CI: use heredoc in wait step to avoid nested-quote issues; remove redundant EBL artifact copy; drop job-level if and unused UNITY_VERSION
* CI: harden wait step (container status check, broader ready patterns, longer timeout); make license return non-blocking
* CI: wait step — confirm bridge readiness via status JSON (unity_port) + host socket probe
* CI: YAML-safe readiness fallback (grep/sed unity_port + bash TCP probe); workflow_dispatch trigger + ASCII step names
* CI: refine license error pattern to ignore benign LicensingClient channel startup; only match true activation/return failures
* Improve Unity bridge wait logic in CI workflow
- Increase timeout from 600s to 900s for Unity startup
- Add 'bound' to readiness pattern to catch more bridge signals
- Refine error detection to focus only on license failures
- Remove non-license error patterns that could cause false failures
- Improve error reporting with descriptive messages
- Fix regex escaping for unity port parsing
- Fix case sensitivity in sed commands
* Add comprehensive Unity workflow improvements
- Add project warm-up step to pre-import Library before bridge startup
- Expand license mounts to capture full Unity config and local-share directories
- Update bridge container to use expanded directory mounts instead of narrow license paths
- Provide ULF licenses in both legacy and standard local-share paths
- Improve EBL activation to capture complete Unity authentication context
- Update verification logic to check full config directories for entitlements
These changes eliminate cold import delays during bridge startup and provide
Unity with all necessary authentication data, reducing edge cases and improving
overall workflow reliability.
* Refine Unity workflow licensing and permissions
- Make EBL verification conditional on ULF presence to allow ULF-only runs
- Remove read-only mounts from warm-up container for Unity user directories
- Align secrets gate with actual licensing requirements (remove UNITY_SERIAL only)
- Keep return-license action at v2 (latest available version)
These changes prevent workflow failures when EBL has issues but ULF is valid,
allow Unity to write preferences during warm-up, and ensure secrets detection
matches the actual licensing logic used by the workflow steps.
* fix workflow YAML parse
* Normalize NL/T JUnit names and robust summary
* Fix Python import syntax in workflow debug step
* Improve prompt clarity for XML test fragment format
- Add detailed XML format requirements with exact specifications
- Emphasize NO prologue, epilogue, code fences, or extra characters
- Add specific instructions for T-D and T-J tests to write fragments immediately
- Include exact XML template and TESTID requirements
- Should fix T-D and T-J test failures in CI by ensuring proper fragment format
* Fix problematic regex substitution in test name canonicalization
- Replace unsafe regex substitution that could create malformed names
- New approach: preserve correctly formatted names, extract titles safely
- Prevents edge cases where double processing could corrupt test names
- Uses proper em dash (—) separator consistently
- More robust handling of various input formats
* CI: NL/T hardening — enforce filename-derived IDs, robust backfill, single-testcase guard; tighten prompt emissions; disallow Bash
* fix: keep file ID when canonicalizing test names
* CI: move Unity Pro license return to teardown after stopping Unity; keep placeholder at original site
* CI: remove revert helper & baseline snapshot; stop creating scripts dir; prompt: standardize T-B validation to level=standard
* CI: remove mini workflow and obsolete NL prompts; redact email in all Unity log dumps
* NL/T prompt: enforce allowed ops, require per-test fragment emission (incl. failures), add T-F..T-J XML templates
* NL suite: enforce strict NL-4 emission; remove brittle relabeling; keep canonicalization + backfill
* NL/T: minimize transcript; tighten NL-4 console reads; add final errors scan in T-J
* ci: add local validate-nlt-coverage helper
* CI: add staged report fragment promotion step (reports/_staging -> reports/) to support multi-edit reporting
* CI: add staged report fragment promotion step (reports/_staging -> reports/) to support multi-edit reporting
* CI: minor polish and guardrails; keep staged reports promotion and placeholder detection
* read_console: default count=50; normalize types str->list; tolerate legacy payload shapes
* read_console: harden response parsing for legacy shapes (data as list, tuple entries)
* Docs: refresh CI workflow and prompts (remove mini suite refs; per-test emissions, staging, guard)
* CI: move T coverage check after staged promotion; accept _staging as present; dedupe promotion step
* CI: make T retry conditional on explicit coverage probe (not failure()); respect _staging in probe
- Windows: Fix path from AppData\Local\Programs\UnityMCP to AppData\Local\UnityMCP
- macOS: Update from /usr/local/bin/UnityMCP to Library/AppSupport/UnityMCP (uses symlink)
- Linux: Change from /home/USERNAME/bin/UnityMCP to .local/share/UnityMCP
- Update Claude Code command examples with correct paths
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove deprecated _trigger_sentinel_async function and all call sites
- Clean up force_sentinel_reload option handling (replaced with Unity's synchronous system)
- Addresses repeated CodeRabbit flagging of unused vestigial code
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
CodeRabbit fixes:
- Fix Python test assertions to use assert instead of print/return
- Update version consistency: server_version.txt from 3.2.0 to 3.3.0
- Assembly definition references already correctly configured
Greptile style fixes:
- Add missing newlines at end of Unity meta files and source files
- Fix test logic assumptions: use GreaterOrEqual instead of exact counts
- Make test assertions more robust for fuzzy matching algorithms
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add nl-unity-suite-full-additive.md: new additive test design that builds state progressively instead of requiring resets
- Update claude-nl-suite.yml workflow to use additive test suite
- Fix validation scoping bugs in ManageScript.cs:
- Correct scoped validation scope calculation (was using newText.Length instead of originalLength)
- Enable always-on final structural validation regardless of relaxed mode
- Unify regex_replace and anchor_insert to use same smart matching logic in manage_script_edits.py
- Additive tests demonstrate better real-world workflow testing and expose interaction bugs between operations
- Self-healing capability: tools can recover from and fix broken file states
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
The TryResolve_PrefersPlayerAssemblies test was failing in CI because it referenced
TicTacToe3D.cs which exists only locally and is not committed to the repository.
- Replace test reference from "TicTacToe3D" to "CustomComponent"
- CustomComponent exists in Assets/Scripts/TestAsmdef/ and is tracked in git
- CustomComponent is properly compiled into the TestAsmdef Player assembly
- Add explicit assertion to verify resolution from correct TestAsmdef assembly
- Test maintains same functionality: verifying ComponentResolver finds user scripts from Player assemblies
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Change from fail-fast to collect-and-continue at component iteration level
- Previously: first component error would halt processing of remaining components
- Now: all components are processed, errors collected and returned together
- Maintain existing collect-and-continue behavior within individual components
- Add comprehensive tests validating the collect-and-continue behavior works correctly
- All valid properties are applied even when invalid ones fail
- Processing continues through exceptions with proper error aggregation
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Performance & Quality Improvements:
- Add shared JsonSerializer to eliminate per-call allocation overhead
- Optimize ComponentResolver with CacheByName for short-name lookups
- Deduplicate Vector3 parsing implementations to reduce maintenance burden
- Improve property name normalization for better fuzzy matching quality
- Reduce log noise by avoiding duplicate component resolution warnings
Code Quality:
- Keep using static import for ComponentResolver (CodeRabbit was incorrect about this)
- Normalize property names consistently in AI suggestions algorithm
- Remove duplicate ParseVector3 implementation
TestAsmdef Fix:
- Revert TestAsmdef back to runtime-compatible (remove "includePlatforms": ["Editor"])
- CustomComponent now works correctly for asmdef testing as originally intended
- Validates that ComponentResolver properly handles both short and FQN for asmdef components
Live Testing Validation:
- All ComponentResolver functionality verified through live MCP connection
- AI property matching working perfectly with natural language input
- Assembly definition support fully functional for both default and custom assemblies
- Error handling provides helpful suggestions and complete context
All 45 C# tests + 31 Python tests still passing. Production ready!
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Critical Bug Fixes:
- Fix operator precedence bug in ManageAsset.cs that could cause null reference exceptions
- Fix GameObject memory leak in primitive creation when name validation fails
- Add proper cleanup with DestroyImmediate when primitive creation fails
ComponentResolver Integration:
- Replace fragile string-based GetComponent() calls with robust ComponentResolver
- Add ComponentResolver integration in ManageAsset.cs for component lookups
- Add fallback to string-based lookup in ManageGameObject.cs for compatibility
Enhanced Error Handling:
- Surface specific ComponentResolver error context in ScriptableObject creation failures
- Add support for setting private [SerializeField] fields in property matching
- Improve debugging with detailed error messages
Assembly Definition Fixes:
- Configure TestAsmdef as Editor-only to prevent build bloat
- Add explicit TestAsmdef reference to test assembly for proper component resolution
- Fix ComponentResolverTests to use accessible CustomComponent instead of TicTacToe3D
Code Quality:
- Disable nullable reference types for legacy codebase to eliminate 100+ warnings
- Maintain backward compatibility while improving reliability
All 45 unit tests pass, ensuring no regressions while significantly improving robustness.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add intelligent property name suggestions when property setting fails
- Implement GetAllComponentProperties to enumerate available properties
- Add rule-based AI algorithm for property name matching (camelCase, spaces, etc.)
- Include comprehensive error messages with suggestions and full property lists
- Add Levenshtein distance calculation for fuzzy string matching
- Cache suggestions to improve performance on repeated queries
- Add comprehensive unit tests (11 tests) covering all ComponentResolver scenarios
- Add InternalsVisibleTo attribute for test access to internal classes
Examples of improved error messages:
- "Max Reach Distance" → "Did you mean: maxReachDistance?"
- Shows all available properties when property not found
- Handles Unity Inspector display names vs actual field names
All tests passing (21/21) including new ComponentResolver test suite.
The system eliminates silent property setting failures and provides
actionable feedback to developers.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Replace Assembly.LoadFrom with already-loaded assembly search
- Prioritize Player assemblies over Editor assemblies using CompilationPipeline
- Support both short names and fully-qualified component names
- Add comprehensive caching and error handling with disambiguation
- Use Undo.AddComponent for proper editor integration
- Handle ReflectionTypeLoadException safely during type enumeration
- Add fallback to TypeCache for comprehensive type discovery in Editor
Fixes component resolution across custom assembly definitions and eliminates
"Could not load the file 'Assembly-CSharp-Editor'" errors.
Tested with:
- Built-in Unity components (Rigidbody, MeshRenderer, AudioSource)
- Custom user scripts in Assembly-CSharp (TicTacToe3D)
- Custom assembly definition components (TestNamespace.CustomComponent)
- Error handling for non-existent components
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
David Sarno
GitHub Actions
Justin Barnett
Jos van der Westhuizen