v0.1.3

Release v0.1.3 — Notes (from CHANGELOG Unreleased)

Added

• No-args primary CLI interactive shell mode (pnpm run cortexa) that starts daemon by default and accepts inline commands.
• Top-level command normalization for --<command> forms and agent alias routing to agents.
• Ingestion scope integration coverage (test:ingestion-scope) validating workspace-scoped chat ingestion and .venv skip behavior.
• Quantized local mini LLM runtime (core/llm/mini-llm.service.ts) with compact n-gram inference and schema-aware JSON completions.
• New primary CLI command surface for local model workflows: llm status, llm train, and llm preview.
• Mini LLM integration coverage (test:llm) to verify train/status/inference and writer/critic/consolidator schema responses.

Changed

• Git release automation now auto-tags v<package-version> on main version bumps via GitHub Actions; tag push continues to build and publish GitHub Releases.
• Ingest command UX now supports optional path and project-id inference from target folder.
• Ingestion pipeline now skips common heavy local/runtime directories (.venv, venv, .cache, target, etc.).
• Copilot chat transcript discovery now prefers current workspace-scoped workspaceStorage before broader root fallback.
• Documentation refreshed across README/runbook/API/contributing to match current CLI and ingestion behavior.
• Evolution writer/critic/consolidator orchestration now uses the local mini LLM client instead of throw-only fallback stubs.

Fixed

• Daemon start now handles already-running instances and EADDRINUSE conflicts without crashing command flow.
• Vector backend outage handling now applies retry cooldown to avoid repeated slow failures while continuing SQLite-first operations.
• Interactive shell exit now returns cleanly (exit path returns code 0) and stops in-process daemon when owned by that shell.

---

This release also includes earlier fixes and improvements referenced in the CHANGELOG, and incremental polish for ingestion, LLM runtime status, and context-stream behavior.