darq v0.1.0

darq takes a GitHub issue and ships the merged PR — autonomously. Plan, implement, review, fix, merge, then Scenario-driven AI Testing (SAT) the result through three persona-driven judges. Every run extracts patterns into a learning store; the next run starts smarter.

Highlights

• 7-step workflow chain — PlanIssue → ImplementIssue → ReviewPr → FixReview → MergePr → SatVerify → LearnUpdate.
• Daemon architecture — long-lived background process owns the SQLite store, event broadcaster, and workflow engine. CLI + TUI are thin clients over a Unix socket.
• ACP-compatible agent runner — default backend opencode; JSON-RPC over stdio.
• SAT engine — persona-based quality gate with junior / senior / maintainer judges, blended score, threshold-gated verdict.
• Learning loop — ruvector + sona_engine extract reusable patterns from every successful run; PlanIssue injects top-K similar patterns on the next run.
• Alien-living TUI — 30 Hz ratatui dashboard with chain topology, agent oscilloscope (idle / thinking / producing), persona dials, event waterfall, and shift-complete card.
• Replay mode — darq tui --replay <jsonl> plays captured event streams through the live TUI.
• Public-ready scaffolding — MIT license, workspace metadata, CI (fmt + clippy + test), CONTRIBUTING, SECURITY, CHANGELOG.

Install

cargo install darq

Requires an ACP-compatible coding agent on your PATH (default: opencode).

Only tested on Linux (Ubuntu 25). macOS and Windows are not supported in v0.1.

Run

darq daemon start                  # one time
darq run issue 42 --full           # plan → implement → review → merge → SAT → learn
darq tui                           # watch live

crates.io

• darq v0.1.0 — CLI + TUI binary
• darq-core v0.1.0 — workflow engine, SAT, learning store

Docs

• README
• Project overview
• Architecture
• Development guide