eagle-hq/EAGLE-4
Speculative decoding drafts plus a harness that tunes speculation to measured costs, for SGLang, vLLM and TokenSpeed.
Why it matters: Up to 3.4x faster single-request decoding on Qwen3.8-27B. Note the source-available licence.
Repository Watchnew on GitHub
dominikmartn/harp
An unofficial Warp fork that runs many Claude Code and Codex agents side by side in one sidebar.
Why it matters: The sidebar flags which agent is waiting on you, which is the real bottleneck in multi-agent work.
alexpiar/be-ambitious-harness
A skill that sets two bars per task, runs a cheap gate check, then uses a verifier blind to the worker's reasoning.
Why it matters: A neat, plain-markdown pattern for pushing agents harder without trusting their own grading.
adityareddy-dev/deskstudio
A design studio that lives inside the coding agent you already use.
Why it matters: Fresh from the agent-skills topic: design iteration without leaving the terminal.
reedodev/explainer-studio
One line in, an explainer video out, in eight hand-drawn styles, as a Claude Code skill.
Why it matters: Shows how far code-rendered video skills have come in a few weeks.
The Orange Wiretop of Hacker News
-
Show HN: SimCity, but with Permits
Satire you can play. The thread turns into a housing-policy debate, as expected.
- Apple broke an essential presentation feature of iPads
- No more icons in every menu by default
- Facebook Pages require paid subscription to post more than two links per month
- What Is Gridifcation
- MicroG: Free-libre re-implementation of Google's proprietary Android user space
- The Real Kevin Bacon
- NixOS is making me less afraid
- Why Windows 11 still ships with Windows 95's memory manager hidden inside it