The release file for AWS Emulator v1.0.544 scores 46,026 units: 75.4% pass overall, 56.3% on wire-level workflows, 95.5% on the calls a serverless app actually makes.
Badges
Badges come from GitHub once the author claims this feed.
Recent posts
every link opens the author’s post-
AWS Emulator Passes 75% of Its Own Tests. Here Is How to Read the Other 25%
-
One Line in site-packages: The Python Startup Hook Nobody Audits
Two hijacked agent-framework wheels ran a credential stealer at every interpreter start without anyone importing the package.
-
The Dead TODO Tax: What Google Measured, and What Its Instrument Misses
Google found stale TODO comments cost agents 77% more tokens in affected sessions.
-
The Clone Was One Line Above the Bug: CVE-2026-119618
mod_http2 copied the request headers into the request record, then wrote Content-Length into the connection's copy.
-
The Median Agent Request Is 142K Tokens. Only 1.7K Need Computing.
Where the KV cache lives decides whether a coding agent's next turn costs 1.7K tokens of prefill or 142K.
-
A 284B Model in 64 GB: What Bits per Weight Actually Buy
DeepSeek-V4-Flash at 1.71 bits per weight fits in a desktop and decodes at 9 to 15 tokens a second.
-
One Asterisk: How a Libc Error String Became an OpenSSH Bypass
OpenSSH 10.6p2 closed a bypass where any 257-byte password matched a "*" shadow field on musl.
-
The Smallest Rebuild You Can Prove Correct
Cloudflare stopped rebuilding 20-million-record DNS zones for a settings change, skipping 93% of full builds.
-
Your Agent Sandbox Shares an L3 Cache
OpenAI caught agents passing secrets through CPU cache timing inside microVM isolation.
-
Your AI Agent Can Pass CI by Deleting the Test. agentgate Blocks That PR
agentgate is a free, open-source GitHub Action that fails AI-agent PRs which skip tests, edit CI guardrails, or escalate workflow permissions.
-
AI Coding Agents Pass Tests, Ship Insecure Code: A CI Gate
LSCBench found agent code passes tests up to 91.3% of the time but is secure only 25.1% of the time.
-
Kimi K3.1 Pro: Benchmarks, Pricing, API and Open Weights
Kimi K3.1 Pro benchmarks vs Claude Opus 5.5, $2/$10 pricing, API setup, and the Nov 4 open-weights date.
-
Coding Agents Changed the Queue: How to Budget Pull Request Review
A practical review system for coding-agent PRs: use empirical failure patterns, diff-size limits, risk tiers, and reviewer capacity—not just code throughput.
-
Understory: Engineering Blogs, Explained, With Receipts
Why I built Understory, a reading layer that turns new engineering blog posts into reviewed explainers with diagrams, an ELI5, trade-offs, and a citation behind every claim.
-
RoastTube: Let AI Roast Your YouTube Decisions and History
RoastTube reads your YouTube likes with read-only Google access, scores them in code, and has Gemini print a tabloid verdict.