0.3.2

CLI/MCP now exposes a light command directory and lets agents request only the operation they need. Intent decisions can be narrowed by platform, provider and action. All 115 commands and the complete contracts remain available.

The packaged ai-app-bridge-use skill keeps the target, observation, action receipt and persisted-result contracts, with detailed documentation loaded on demand. Duplicate schema unions are flattened without changing the execution contracts.

All consumer versions are aligned to 0.3.2: CLI/MCP, Android SDK and Gradle plugin, native iOS Git tag, Flutter on pub.dev, and the Web npm package. Android/iOS/Flutter/Web execution behavior is unchanged in this patch.

Validation includes a fresh npm tarball installation with native compilation and CLI/MCP protocol checks, public Android JitPack artifacts, Flutter publication checks, and a Reader real-device reading workflow. The root SPM iOS Simulator build passed on the 0.3.1 source; the 0.3.2 native iOS source changes only the SDK version marker.

Reader Intent observations were turned into a Script. With both the globally installed CLI/MCP and the installed Reader Android SDK at 0.3.2, the final run completed 46 calls and 23 device assertions in 23.337 seconds, with no failed or inconclusive assertions. Reading, chapter selection, settings and reopening the book were verified. Mobile events and the final result were persisted; the evidence archive passed integrity verification and includes four screenshots. This validates the selected reading workflow; full-app regression remains outside this release check.

The failed 0.3.1 JitPack build record remains available. A fresh 0.3.2 build succeeded with the original performance thresholds and runtime implementation unchanged.