nyldn/claude-octopus v9.4.0

v9.4.0 — Four-Way Debates, Auto Code Review, Wave 1 Decomposition

on GitHub

What's New

Four-Way AI Debates

Sonnet joins as a permanent 4th participant alongside Codex, Gemini, and Claude/Opus. Runs in parallel via Agent(model: "sonnet") — no added latency, no extra cost. Each model gets a distinct role: Codex (technical critic), Gemini (strategic thinker), Sonnet (pragmatic implementer), Opus (moderator).

Auto Code Review + E2E Verification

After any /octo:develop, /octo:embrace, or /octo:deliver workflow, two Sonnet agents automatically launch in parallel — one code reviewer, one E2E tester. Findings are presented before the "what next?" prompt. No manual request needed.

Wave 1 Decomposition

First step in the orchestrate.sh decomposition plan. Three new lib modules extracted (22,668 → 22,377 lines):

• lib/utils.sh — JSON parsing, sanitization, validation utilities
• lib/similarity.sh — Text similarity and convergence detection
• lib/models.sh — Model catalog and capability lookup

Plus a monolith guard test that prevents orchestrate.sh from growing back.

9 Review Pipeline Bug Fixes

The entire /octo:review pipeline was silently dropping findings due to 9 interlocking bugs (#182-#190). All fixed — review findings now flow through to the user correctly.

Also Fixed

• list_models --tier parsing bug (pre-existing)
• log() forward-reference in extracted utils
• validate_output_file silent failure when RESULTS_DIR unset
• Test patterns updated for v9.3.1 model flag changes

Full Changelog: v9.3.0...v9.4.0