Opus 5.5 Sets Strategy, Sol 6.2 Writes Code: A Multi-Agent Workflow
PawelHuryn · x · 2026-11-19
Pawel Huryn shares his current multi-model agent workflow and has open-sourced 45 agent skills alongside it (GitHub repo product-engineering, 653 stars).
- Division of labor: Opus 5.5 xhigh handles judgment, common sense, strategy, and managing other agents; Sol 6.2 xhigh takes care of implementation and the code review loop
- Open-source skills: drawn from hands-on experience shipping three real products, they cover the full pipeline of "decide what to build → build with Claude Code/Codex → have a second model review → launch," including commands like /delegation, /codex-agents, /code-review, and /independent-review
- A gripe: driving the Antigravity CLI with Neon would violate its terms of service, so he skipped that setup
More from coding & agent
- DataTalks.Club launches free 5-week course on GPU-less generative media agents — Al_Grigor · 2026-11-19
- Codex refuses to write security advisories for bugs it found itself, dev says — matthew_d_green · 2026-11-19
- Claude-powered guide maps 6 GTA 6 money routes as pre-orders top $1.5B — socialwithaayan · 2026-11-19
- Booking travel remains the hardest agent task, dev falls back to Mac-local setup — lucasmeijer · 2026-11-19
- ElevenLabs launches Scribe v3 Live Medical with 10,000-entry custom term lists — HeyAmit_ · 2026-11-19
- Developer on why personal AI agents still fail: 1Password auth, Mac control, mobile use — lucasmeijer · 2026-11-19