system1-agents 0.2.0

Second release. New decision backends, bounded recovery on browser and desktop, desktop text/input polish, Public231 benchmarks, and a snake recipe — on top of the 0.1.0 nine-agent CLI/MCP base.

Install

git clone https://github.com/ThinkFlowLab/system1-agents && cd system1-agents
uv sync && cp .env.example .env

From Claude Code, Codex or Cursor: npx skills add ThinkFlowLab/system1-agents. Claude Code plugin: claude plugin marketplace add ThinkFlowLab/system1-agents && claude plugin install s1a@system1-agents.

Highlights since 0.1.0

Decision models

• --model clm: CLM via /v1/systemone (no torch / no cloud key on this side)
• --model laya-served: Laya over HTTP from system1-omni (or plain laya-serve)
• --model omnijev: OmniJev vision-language decisions in process (screenshot ticks)
• Optional local Cua-S1 4B (cua-four-b / CUA_S1_VARIANT=4b)
• Faster Laya load (~3s vs ~35s); browser-state compaction for Laya's context window
• MCP decide accepts model="jev"|"laya"|"cua"

Agents & recovery

• Bounded recovery on browser (--rethink) and desktop (shared attempt/timeout knobs)
• Verified desktop text input (--text / --text-target / --text-mode) plus pixel/window targeting
• Snake recipe + game client with recorded evidence
• Browser runs save final.png; image-capable models get viewport PNGs per tick

Benchmarks & tooling

• Public231 decision-model benchmark (HTTP evidence, quality metrics, latency)
• Google Flights driver comparison rerun archived under docs/results/
• Agent recipe index + authoring skill; CodeRabbit + review-pr / merge-readiness gates
• Windows core CI on main/schedule; rail/browser token billing fixes

CHANGELOG.md (Unreleased section) has the full list.