nyldn/claude-octopus v8.54.0

v8.54.0 — Multi-agentic /octo:research

on GitHub

Multi-agentic /octo:research — parallel Agent dispatch replaces single Bash subprocess

Changed

• Multi-agentic /octo:research — Refactored from single Bash(orchestrate.sh probe) call (120s timeout) to parallel Agent(run_in_background=true) subagents. Each perspective calls orchestrate.sh probe-single independently — no timeout constraint. Claude synthesizes in-conversation instead of Gemini synthesis that frequently timed out.
• User-configurable research intensity — /octo:research and /octo:discover now ask intensity before launching: Quick (2 agents, 1-2 min), Standard (4-5 agents, 2-4 min), Deep (6-7 agents with web search, 3-6 min).
• Gemini-first launch ordering — Higher-latency Gemini agents launch first, reducing total wall-clock time.

Added

• probe_single_agent() + probe-single dispatch command in orchestrate.sh
• test-probe-single.sh — 26 tests covering function, dispatch, integration, and backward compatibility

Fixed

• test-knowledge-routing.sh — Pre-existing SIGPIPE flake (grep -q → grep -c)

Backward Compatible

• probe_discover() untouched — /octo:embrace still uses the original single-subprocess path
• discover|research|probe dispatch case unchanged