v6.4.2
vllm-project/vllm-omniv6.4.2Sep 25, 2026by obra
Upgrading?
AI Summary
Optimizes the 'writing-plans' skill to generate leaner, faster plans by recording decisions (signatures, assertions) rather than full code transcripts, significantly improving speed and token efficiency.
Key Highlights
- Plans now record decisions (signatures, test assertions) instead of full code.
- Self-review checks plan length proportion to spec to prevent transcript behavior.
- Plans executed successfully on planted-defect probes with improved performance (quarter time, third tokens).
- Removed unused documentation files (`plan-document-reviewer-prompt.md`, `CLAUDE.md`).
New Features
- New plan writing logic focusing on decisions.
- Improved performance for plan generation.
- Updated documentation structure.
Full Release Notes
`writing-plans` produces leaner plans, faster. Plans now record the decisions an implementer needs (signatures, test assertions, the spec's values) instead of writing out the code. Some frontier models, including Opus 5.5, could get overzealous during plan writing and, with certain prompting, would sometimes try to implement the entire project while designing the plan. The new skill keeps planning focused on the plan. When we reproduced the original report, the scratch builds went away, and plans took a quarter of the time and about a third of the tokens. Thanks to Harper Reed for the report and session bundle. (#2333) ### Writing Plans - **A plan records decisions. It's not a transcript of the code.** "What a Step Contains" replaces the "No Placeholders" section. A test step names the test and its assertions. A code step gives the exact signature, the file, and the spec's values, and includes a body only for an algorithm those don't determine. A verification step gives the command and its passing output. A reference to another task goes through that task's Interfaces block. Placeholders are still called out as the opposite failure. (#2333) - **Self-review checks proportion.** The plan compares its own length to the spec's. A plan several times longer than the spec is a transcript, and when code blocks dominate, bodies get replaced with signatures and test assertions. (#2333) - **The plan's reader is described as capable:** an engineer who writes idiomatic code once they know the exact interface and test. This replaces "zero context, questionable taste." Steps are now sized as "one action with a checkable result" instead of "2-5 minutes." (#2333) - Every plan written by the new skill executed 9/9 against planted-defect probes on Sonnet 5, the same result as full-code plans. (#2333) - Removed `plan-document-reviewer-prompt.md`. Nothing referenced it. (#2333) ### Documentation - Removed `CLAUDE.md`. Claude Code now reads `AGENTS.md` directly, but only when no `CLAUDE.md` exists, so keeping the one-line pointer would have hidden the real guidelines.
Related Repositories
public-apis/public-apis★ 491,135
A collective list of free APIs
NousResearch/hermes-agent★ 255,963
The agent that grows with you
yt-dlp/yt-dlp★ 200,054
A feature-rich command-line audio/video downloader
microsoft/markitdown★ 191,491
Python tool for converting files and office documents to Markdown.
Significant-Gravitas/AutoGPT★ 187,703
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.