- config: point telemetry_endpoint to Cloud Run default
- telemetry: log effective endpoint/timeout; reject localhost endpoints
- server: telemetry logger at normal level with rotating file; default timeout=5s when unset
- Logging to stderr with force; quiet httpx/urllib3
- Async lifespan fix; defer telemetry in first second
- Bounded telemetry queue with single worker
- Reduce initial Unity connect timeout to 1s
- Keep server_version in file
- Detect and clean up legacy server installations in LocalApplicationData
- Prevents accumulation of old server copies during package updates
- Improves cleanup of Windows-specific legacy installation paths
- README path separators (todo in separate doc commit)
- manage_gameobject: pop prefabFolder not prefab_folder
- execute_menu_item: make sync to avoid blocking event loop
- telemetry: validate endpoint scheme (allow http/https only) and re-validate at send time
- 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>
GitHub Actions
dsarno
Justin Barnett