[AfonsoNeto]

Summary

Brings the gem to v0.2.0 with a focus on making the local-LLM integration accurate and reliable.

API layer

• Switch from Ollama's proprietary /api/generate to the standard OpenAI-compatible /v1/chat/completions endpoint — now genuinely works with Ollama, LM Studio, llama.cpp llama-server and vLLM
• URL normalization: accepts base URLs with or without /v1 and trailing slashes
• Enforce TLS with certificate verification for https endpoints
• HTTP timeouts (5s open / 120s read), HTTP status checks and typed errors (ConnectionError, ApiError) with friendly messages
• Strip ANSI escape sequences from model output (prevents terminal escape injection via prompt-injected diffs)
• 2048-token budget so thinking models (qwen3) finish reasoning before answering; inline <think> blocks stripped
• Output cleaning: code fences, quotes and multi-suggestion replies reduced to a single 72-char subject line

CLI

• commit now commits by default after a y/n confirmation (closed stdin counts as "no" so non-interactive runs never auto-commit)
• New --message-only flag replaces print-only behavior; --conventional for Conventional Commits format; --length for subject length
• New models command lists models from the server; config picks from that list when reachable
• Fixed options.length collision that truncated every generated message to 3-4 characters
• Fixed option resolution where explicitly passed --url/--model were ignored
• Skip the LLM call when nothing is staged; cap diff size at 12k chars; handle detached HEAD

Housekeeping

• Default model → qwen3:8b
• Minitest suite (23 tests) with a dependency-free mock HTTP server
• GitHub Actions CI (Ruby 3.1–3.4) with tests and bundler-audit dependency scanning
• MIT license, gemspec cleanup, README rewrite with a terminal demo GIF (regenerable via docs/generate_demo.py)

Test plan

• bundle exec rake test — 23 tests, all green (verified repeatedly, plus on CI across Ruby 3.1–3.4)
• Live end-to-end runs against local Ollama (qwen3:8b and qwen2.5-coder:7b): generation, confirmation flow, EOF handling and --message-only