v0.6.3

Additive release: ships two optional shell-script examples that wrap the existing nlink-jp RAG CLIs (gem-rag for Vertex AI Gemini, lite-rag for local LLM via OpenAI-compatible API) so an agent session can ask questions against a pre-indexed knowledge base. No core code changes.

Added

• examples/search-kb-gem.sh — wraps gem-rag ask --json and surfaces it as the search-kb-gem agent tool. Requires gem-rag installed, Vertex AI ADC, and a corpus pre-indexed via gem-rag index --dir <docs>.
• examples/search-kb-lite.sh — wraps lite-rag ask --json and surfaces it as the search-kb-lite agent tool. Requires lite-rag installed, ~/.config/lite-rag/config.toml, a running local LLM endpoint (e.g. LM Studio), and a corpus pre-indexed via lite-rag index --dir <docs>.
• Both scripts use @category: read, @timeout: 120 (RAG round-trips routinely exceed the 30s default), pass the user query through unchanged, and emit a structured JSON error when the backend CLI is not on PATH so the agent can guide the user to install it.
• bundled_test.go: TestExamples_AreReadableAndHaveToolHeader guards all four examples (existing web-search / generate-image plus the new pair) against header-stripping refactors.

Compatibility

• Examples remain opt-in: bundled.Install skips the examples/ subdirectory, so existing users see no change until they manually copy a script into their tool dir.
• No persistence-format changes, no API changes, no behavioural changes for sessions that don't enable the new tools.

Install

Download shell-agent-v2-v0.6.3-darwin-arm64.zip, unzip, and move shell-agent-v2.app to /Applications/.