## (daily refresh)

- Added unreal-agent (harnesses), Unreal Labs' MIT Go async-first harness, 1,879 stars at day three and v0.2.0, extending the harness matrix to 28 columns [glm-5.3-flash]
- Rejected AnyJev (zero HN footprint), Whiteboard (days-old IDE), magpie (gateways question), mikehasa/golive-skill (community skill), Step Code (four months quiet), jev-chat apps, and the sub-bar tail [glm-5.3-flash]
- Corrected vibe-kanban's release record (a timestamped v0.1.45 prerelease exists, npm still 0.1.44) and emdash's release-gap claim to three days [glm-5.3-flash]
- knowhere: replaced the $5 signup credit with a 14-day trial (Pricing sentence and Bottom line fixed); ctx: withdrew the pro subscription (Pricing rewritten, Price history withdrawal row) [glm-5.3-flash]
- jevbench: materially revised with v1.4.0-v1.4.2 (308 sealed decisions, a public-to-sealed gap penalty; Jev first at 63.3, JevK5 and Hopper new at #2 and #3, SemIf 73.1 to 47.7) [glm-5.3-flash]
- opensandbox: shipped its first stable 1.1.0 umbrella release; openai-for-science: rewritten for the public Navier-Stokes Lean formalization and the Buckmaster-Alpoge Euler certificates [glm-5.3-flash]
- model-selection: added GPT-6 Sol and Luna columns (Sol $2/$10, Luna $0.10/$0.50) and re-grounded the Kimi rows on Moonshot's official prices (kimi-k2.7-code $0.95/$4.00 cache $0.19, kimi-k3 $3/$15 cache $0.30); corrected GLM-5.3-FlashX's cached-read cell to $0.090 and fixed the transposed context-behavior cells [glm-5.3-flash]
- Removed the automated-research matrix's stale verification preamble and converted its plain-text headers to member links [glm-5.3-flash]

## (daily refresh, same-day pass)

- Fixed title-order violations in the executions, protocols, and skills index pages and matrices (GitHub Agentic Workflows before GitHub Copilot automations; ACP, Agent Host Protocol, and AG-UI before A2A and AGENTS.md; Agent Skills open standard before Agent-Native), no cell content moved [glm-5.3-flash]
- antigravity: gained the required Price history table seeded with the 2026-08-24 individuals baseline ($0/month) [glm-5.3-flash]
- Rejected bestvaluemodel (a one-day-old repo with 10 stars, no license, deriving entirely from member Artificial Analysis) and llm-list.com (fully derivative of models.dev, OpenRouter, Arena, Artificial Analysis, and Hugging Face) [glm-5.3-flash]
- Added the Di Zhang RLCD deconstruction to the jev note's Status and References; MoJev stays uncategorized (29 stars, self-published) [glm-5.3-flash]