Devin’s maker shipped SWE-2, scoring 50.0% on its FrontierCode 1.1 Main benchmark—within a point of Anthropic’s Fable 5.1—while claiming ~64% lower cost. It’s live in Devin Desktop and CLI now, with Web and Fusion rolling out; no standalone API or weights.
Shown on HN Oct 26, the tool runs a shared counting semaphore so OpenCode, Codex CLI, Claude Code and other agents across repos or worktrees take turns on heavy jobs like cargo test while queued agents keep coding, with permits freed when the owning process exits. Install with npm install g @aron98/load lock then load lock install harness claude code (or codex, opencode, cursor); it's Linux first and the Claude Code and Cursor hooks aren't live verified yet…
Shown on HN Oct 26, the free Rust CLI swaps lsof | xargs kill pipelines for commands like proc why :3000 and proc free :3000 , with json on every command, fixed exit codes and dry run before anything changes. An agent skill ( npx skills add yazeed/proc cli ) teaches Claude Code, Codex, Cursor and Gemini CLI to clean up their own dev servers and leave processes they didn't start alone (site)…
Shown on HN Oct 26, it runs three.js plus manifold 3d in a tab for boolean and mesh edits, and gives your agent a render preview tool that takes front, right, top, iso and custom angle screenshots so it can check the geometry and fix it. The author pitches it as a shareable alternative to installing Blender and blender mcp, and says it's a test of an agent's spatial reasoning rather than a high detail mesh generator (try it)…
Shown on HN Oct 26, the MIT licensed agent runs as one SBCL process and its model's single tool evaluates Lisp in that running image, so anything you ask it to learn (say, a read pdf function) runs at once, is saved as a git committed file and reloads after a restart. Your terminal, a web tab, your phone and Discord, Telegram or Slack bots all attach to the same process, so a tool written in one shows up in all of them (docs)…
Agents open, write and land tasks through MCP tools, and each landing is tested against the exact current head of main and replayed if main moved, while still writing ordinary Git commits so GitHub and deploys keep working. In Murmell's own benchmark with 8 Claude Code agents, main went red 0 of 102 times versus 14 of 104 under Git; try it with avp up in a checkout (code)…
Spotted something we missed? Start a thread.