Release v0.8.2: Code review fixes, documentation restructure, VS Code extension improvements

Comprehensive code review and documentation overhaul. Two full review
cycles identified and fixed 40+ issues across math, architecture, domain
correctness, and documentation.

Bug fixes:
- AnalyzeSwerveTool.poseDistance always returned zero (flat struct layout)
- MoI R² used inconsistent torque formula between OLS and residual loops
- time_correlate included NaN/Infinity, used single-signal quality score
- CSV export struct columns didn't match header ordering
- 2024 Crescendo auto amp scoring was 5 instead of 2
- HTTP SSE committed 200 before executor acceptance check
- Cross-correlation center lag could cause ArrayIndexOutOfBounds
- DataQuality gaps changed from count-based to duration-based ratio
- TbaClient.apiKey not volatile for HTTP thread visibility

VS Code extension:
- Generates .mcp.json on activation for Claude Code compatibility
- Passes settings as both CLI args and env vars
- Auto-detects WPILib IDE JDK (matches season)
- Added extension icon
- New buildExtension Gradle task (package without installing)

Documentation:
- README restructured: Installation first, split detailed docs into
  vscode-extension/README.md, doc/STANDALONE.md, doc/DEVELOPMENT.md
- AI model capability caveats added across all docs
- TOOLS.md confidence levels fixed (moderate→medium, added insufficient)
- Default team number set to 2363

Stress tests:
- Synthesize defaults when no config file present
- SLF4J logging reduced to warn, JUnit tree output, console forwarding

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: caparomula

.github/workflows/ci.yml

@@ -0,0 +1,48 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  build:
+    strategy:
+      matrix:
+        os: [ubuntu-latest, windows-latest]
+    runs-on: ${{ matrix.os }}
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up JDK 17
+        uses: actions/setup-java@v4
+        with:
+          java-version: '17'
+          distribution: 'temurin'
+
+      - name: Setup Gradle
+        uses: gradle/actions/setup-gradle@v4
+
+      - name: Build and test
+        run: ./gradlew test shadowJar
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '20'
+
+      - name: Check VS Code extension compiles
+        working-directory: vscode-extension
+        run: |
+          npm ci
+          npm run compile
+
+      - name: Upload shadow JAR
+        if: matrix.os == 'ubuntu-latest'
+        uses: actions/upload-artifact@v4
+        with:
+          name: wpilog-mcp-jar
+          path: build/libs/wpilog-mcp-*-all.jar
+          retention-days: 14

.github/workflows/release.yml

@@ -0,0 +1,55 @@
+name: Release
+
+on:
+  push:
+    tags: ['v*']
+
+permissions:
+  contents: write
+
+jobs:
+  release:
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up JDK 17
+        uses: actions/setup-java@v4
+        with:
+          java-version: '17'
+          distribution: 'temurin'
+
+      - name: Setup Gradle
+        uses: gradle/actions/setup-gradle@v4
+
+      - name: Build shadow JAR
+        run: ./gradlew shadowJar
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '20'
+
+      - name: Bundle JAR into extension
+        run: |
+          cp build/libs/wpilog-mcp-*-all.jar vscode-extension/server/wpilog-mcp-all.jar
+
+      - name: Build VS Code extension
+        working-directory: vscode-extension
+        run: |
+          npm ci
+          npm run compile
+          npx @vscode/vsce package
+
+      - name: Extract version from tag
+        id: version
+        run: echo "version=${GITHUB_REF_NAME#v}" >> "$GITHUB_OUTPUT"
+
+      - name: Create GitHub Release
+        uses: softprops/action-gh-release@v2
+        with:
+          generate_release_notes: true
+          files: |
+            build/libs/wpilog-mcp-${{ steps.version.outputs.version }}-all.jar
+            vscode-extension/*.vsix

.gitignore

@@ -5,6 +5,61 @@ All notable changes to wpilog-mcp will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.8.2] - 2026-03-26
+
+### Added
+- **YAML configuration support** — `ConfigLoader` now parses `servers.yaml` with 3-layer defaults merging (top-level defaults, per-server overrides, environment variable interpolation).
+- **VS Code extension** — `wpilog-analyzer` extension with `McpServerDefinitionProvider` registration and `.mcp.json` generation for Claude Code compatibility. Auto-detects WPILib IDE JDK, log directory, and bundled JAR. Extension icon added.
+- **VS Code extension `.mcp.json` generation** — Extension writes `.mcp.json` on activation with all settings (log directory, team number, TBA key) so Claude Code can discover the server.
+- **Standalone installer scripts** — `install.ps1` (Windows) and `install.sh` (macOS/Linux) for one-line installation outside VS Code.
+- **CI/CD workflows** — GitHub Actions for build, test, and release automation.
+- **`buildExtension` Gradle task** — Builds the server JAR, compiles TypeScript, and packages the `.vsix` without installing. Complements `bundleExtension` (JAR only) and `installExtension` (build + install).
+- **Stress test default configuration** — Stress tests now synthesize defaults (`~/riologs`, team 2363, TBA from `TBA_API_KEY` env var) when no config file is present. No `servers.yaml` entry required.
+
+### Fixed
+- **`AnalyzeSwerveTool.poseDistance` always returned zero** — Odometry drift analysis now handles flat struct layout (`{x, y}`) from Pose decoders, not just nested `{translation: {x, y}}`.
+- **MoI R² computation used inconsistent torque formula** — Residual loop now uses the same torque sign convention as the OLS fit.
+- **`time_correlate` included NaN/Infinity values** — Filter now requires `Double.isFinite()`. Uses worst-of-two quality scores.
+- **`FrcDomainTools.extractTranslation` only handled nested layout** — Now supports both nested `{translation: {x, y}}` and flat `{x, y}` from struct decoders.
+- **CSV export struct column mismatch** — Explicit field ordering for Pose2d, Pose3d, SwerveModuleState matches header row. Generic structs use deterministic alphabetical ordering.
+- **2024 Crescendo auto amp scoring** — Corrected from 5 to 2 points per game manual.
+- **TBA quarterfinal match key** — Added explicit `qf` comp level handling.
+- **HTTP SSE response committed before executor check** — Moved `sendResponseHeaders(200)` inside the executor task so `RejectedExecutionException` returns 503.
+- **LogManager shutdown lifecycle** — `shutdownNow()` + `awaitTermination()` prevents resource leaks. Disk cache shutdown waits for sync executor.
+- **Cross-correlation center lag guard** — Returns `FAILED` when center lag exceeds array bounds, preventing `ArrayIndexOutOfBoundsException`.
+- **`DataQuality` gap detection** — Changed from count-based to duration-based gap ratio for confidence level calculation.
+- **DS timeline linear scan** — Replaced with binary search (`findFirstIndexAtOrAfter`) for teleop deferred emit.
+- **Vision quality fallback** — Falls back to pose entry quality when no target entries found.
+- **Overshoot detection absolute minimum** — Added `Math.01` minimum threshold to prevent false positives near zero setpoints.
+- **Median calculation for even-length arrays** — Now averages two middle elements.
+- **Battery voltage off-by-one** — Changed `voltageValues.size() - 10` to `voltageValues.size() - 1`.
+- **`TbaClient.apiKey` not volatile** — Added `volatile` for safe publication to HTTP handler threads.
+- **`health_check` misleading field name** — Renamed `cache_memory_mb` to `jvm_heap_used_mb` to accurately reflect that it measures total JVM heap, not cache-specific memory.
+- **Main.java help text** — Changed stale `servers.json` reference to `servers.yaml`.
+- **CHANGELOG gap threshold** — Corrected "3x-median" to "5x-median" to match code.
+- **Launcher scripts hardcoded WPILib year** — Now dynamically scan for latest installed year.
+- **Unix launcher missing JAVA_HOME fallback** — Added between WPILib scan and bare `java`.
+- **CI extension compile restricted to Ubuntu** — Removed `if: matrix.os == 'ubuntu-latest'` from Node.js/extension steps.
+- **`ExportTools` missing parameter validation** — `export_csv` now uses `getRequiredString()` for the `name` parameter.
+- **Correlation guidance language** — Softened wording for moderate correlations.
+- **CHANGELOG `WPILOG_BIND`** — Corrected to `WPILOG_HTTP_BIND`.
+
+### Changed
+- **Documentation restructured** — README slimmed to focus on installation and features. Detailed docs split into `vscode-extension/README.md` (extension settings, upgrading, uninstalling), `doc/STANDALONE.md` (standalone install, configuration, Docker), and `doc/DEVELOPMENT.md` (building, project structure, contributing).
+- **Installation section leads README** — VS Code extension and standalone install presented as two equal paths with links to dedicated docs. Notes that both can coexist independently.
+- **AI model capability caveats** — Added notes across documentation that analysis quality depends on the AI model used.
+- **Default team number** — Changed from 0 to 2363 in generated config templates, extension defaults, and stress test fallbacks.
+- **TOOLS.md confidence levels** — Fixed "moderate" to "medium", added missing "insufficient" level, corrected `servers.json` to `servers.yaml`.
+- **`analyze_can_bus` categorization** — Moved from Robot Analysis to FRC Domain in README tool table to match code module.
+- **Stress test output** — SLF4J logging reduced from `debug` to `warn`, JUnit output uses compact `tree` mode, stdout/stderr forwarded to console.
+
+## [0.8.1] - 2026-03-24
+
+### Added
+- **Configurable HTTP bind address** — New `bind` config field / `-bind` CLI flag / `WPILOG_HTTP_BIND` env var controls which network interface the HTTP transport listens on.
+- **Configurable endpoint path** — New `endpoint` config field allows customizing the HTTP endpoint path.
+- **Origin allowlist** — New `origins` config field restricts CORS access to a configurable list of allowed origins.
+
 ## [0.8.0] - 2026-03-24
 
 ### Added
@@ -208,7 +263,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 #### LLM Epistemological Guardrails
 - **Trojan Horse tool descriptions (§6.1)** — All 20 analytical tools now embed interpretation guidance in their MCP `description()` strings. Five guidance constants in `ToolUtils` (`GUIDANCE_UNIVERSAL`, `GUIDANCE_STATISTICAL`, `GUIDANCE_POWER`, `GUIDANCE_MECHANISM`, `GUIDANCE_MATCH_ANALYSIS`) provide consistent, category-appropriate caveats about single-match limitations, sample size uncertainty, correlation-vs-causation, and alternative explanations. Informational tools (`list_entries`, `read_entry`, etc.) are unchanged.
-- **Data quality metadata (§6.5)** — New `DataQuality` record computes quality metrics from any `List<TimestampedValue>`: sample count, time span, gap count/max (adaptive 3x-median threshold), NaN/Infinity count, effective sample rate, timing jitter, and a composite quality score (0.0–1.0). `ResponseBuilder.addDataQuality()` serializes these into a `data_quality` JSON object and auto-warns when score < 0.5. Integrated into `get_statistics` as reference implementation.
+- **Data quality metadata (§6.5)** — New `DataQuality` record computes quality metrics from any `List<TimestampedValue>`: sample count, time span, gap count/max (adaptive 5x-median threshold), NaN/Infinity count, effective sample rate, timing jitter, and a composite quality score (0.0–1.0). `ResponseBuilder.addDataQuality()` serializes these into a `data_quality` JSON object and auto-warns when score < 0.5. Integrated into `get_statistics` as reference implementation.
 - **Output contextual framing (§6.2)** — New `AnalysisDirectives` class generates `server_analysis_directives` in tool responses. `fromQuality(DataQuality)` factory auto-generates guidance from detected issues (low sample count, gaps, NaN, short time span). Builder methods `addGuidance()`, `addFollowup()`, and `addSingleMatchCaveat()` allow tool-specific enrichment. `ResponseBuilder.addDirectives()` serializes into `confidence_level`, `sample_context`, `interpretation_guidance[]`, and `suggested_followup[]` fields.
 
 ### Fixed