Apache Polaris 1.7.0 fixes retry bugs that silently corrupt Iceberg tables. Idempotent writes and stricter credential vending make catalog operations safe over unreliable networks. Upgrade to prevent data loss without touching query…

Apache Polaris 1.7.0 and the Quiet Work of Making a Catalog Trustworthy

Four Signals — The Wire

foursignals.dev

Astro's GitHub backlog dropped from 200+ to ~20 issues using a multi-agent AI triage pipeline. Cloudflare just open-sourced it as triagebot-action.

Astro’s GitHub issue backlog is heading to zero for the first time in 5 years. Now Cloudflare is open-sourcing the tool that did it.

Four Signals — The Wire

foursignals.dev

addyosmani/agent-skills ships 24 production-grade engineering skills as slash commands for 70+ AI coding agents. Install with `npx skills add`.

[GitHub Trending] addyosmani/agent-skills

Four Signals — The Wire

foursignals.dev

DeepSeek-Reasonix ships as a single static Go binary with plugin-driven AI coding for the terminal.

[GitHub Trending] esengine/DeepSeek-Reasonix

Four Signals — The Wire

foursignals.dev

LoopX open-sources a durable state kernel for multi-agent workflows. 200+ hour trajectories across Codex, Claude Code, Cursor.

[GitHub Trending] huangruiteng/loopx

Four Signals — The Wire

foursignals.dev

Claude refused 41.3% of Stripe API tasks (62/150) but 0% on Zod controls. Empty refusals compiled cleanly, masking failures. Payment-intent creation refused 10/10 times; refunds only 1/10. Add empty-response detection to AI code pipelines.

I type-check AI-generated SDK code against the real package. Claude refused a third of my Stripe tasks.

Four Signals — The Wire

foursignals.dev

OpenCode, an open-source Go coding agent, now supports 75+ LLM providers and 160K GitHub stars, surpassing Claude Code. Its client/server architecture fights vendor lock-in.

OpenCode: The Open Source Coding Agent That Doesn't Lock You In 🔓

Four Signals — The Wire

foursignals.dev

JetBrains previews LSP extension bringing IntelliJ's Java/Kotlin intelligence to VS Code, Cursor, and Claude Code agents.

IntelliJ IDEA Goes LSP: Java and Kotlin Intelligence Comes to VS Code, Cursor, and Agentic Flows

Four Signals — The Wire

foursignals.dev

LLMs amplify expert advantage more than novice gain. Domain experts extract higher value from LLM outputs. Invest in prompt engineering training for your top engineers, not just broad rollouts.

Alibaba's Qwen3.8-Max autonomously coded for 16 days—265 commits, 127 PRs. At $2/$6 per million tokens, it demands mature orchestration and guardrails for long-running agents.

Alibaba’s AI coded for 16 days straight and every commit is on GitHub

Four Signals — The Wire

foursignals.dev

LLMs nail pixel-perfect layouts in HTML/CSS. Wire an HTML-to-image API over MCP (e.g., html2img.com) so agents like Claude Code render deterministic OG cards and invoices—no diffusion lottery.

Your AI agent can't design images. It can write HTML.

Four Signals — The Wire

foursignals.dev

Deciding between skill or sub-agent? Azure's Pattabiraman has 4 dimensions: iteration, voice, human gate, frequency. Skills reuse; sub-agents avoid context pollution.

Azure and Community Guidelines on Choosing Between a Skill or a Sub-Agent

Four Signals — The Wire

foursignals.dev

JSON silently truncates floats and coerces types. Use JSON Schema or Pydantic to catch corruption before it spreads.

Your JSON Is Lying to You

Four Signals — The Wire

foursignals.dev

SQLite's 100% branch coverage and 1,000+ fault-injection tests offer reliability lessons for your edge services.

Skip OCR for 54% of PDFs with Firecrawl's pdf-inspector – pure Rust, under 200ms, 0.875 benchmark. Supports Python, Node.js, WASM.

[GitHub Trending] firecrawl/pdf-inspector

Four Signals — The Wire

foursignals.dev

Docker proves your app runs locally. Kubernetes proves it's operable—by forcing answers to four questions: DNS name, graceful shutdown, readiness probes, stateless identity.

Docker Proves Your App Runs. Kubernetes Proves It's Operable.

Four Signals — The Wire

foursignals.dev

MiniMax H3 lands in ComfyUI with open weights. Generates 2K video + stereo audio for 15 seconds, runs locally on a 3060 GPU. Multimodal context, motion transfer, reference-to-video editing.

MiniMax H3 Day-0 Support in ComfyUI: Open Weights, Native Audio, and 2K Video

Four Signals — The Wire

foursignals.dev

TypeScript 7 rewrites compiler in Go—10x faster. VS Code from 125.7s to 10.6s, project load 9.6s→1.2s, memory -18-26%. Upgrade and tune --checkers to CI cores.

TypeScript 7 Went Native: What Actually Changes And What Doesn't

Four Signals — The Wire

foursignals.dev

Stream 80B Qwen on a Mac using just 4.3GB RAM—Swiftlet runs MoE models via SSD streaming, hitting 4-5 tok/s on M5.

Show HN: Run an 80B Qwen in 4.3 GB of RAM on a Mac, and a 35B on an iPhone

Four Signals — The Wire

foursignals.dev

AWS signed a multiyear deal with Superblocks to embed vibe-coding in private clouds via Aurora and Bedrock. Open models now 29% of Vercel's AI traffic. Single-model bets risk getting fired.

AWS is helping vibe-coding startup Superblocks, and the implications are big

Four Signals — The Wire

foursignals.dev

Run 70B LLMs on a 4GB GPU? AirLLM streams layers, not full models. Now supports Kimi K3 (2.8T) on 3.72GB and DeepSeek-V3 (671B) on ~12GB. No quantization needed.

[GitHub Trending] lyogavin/airllm

Four Signals — The Wire

foursignals.dev

DeepSeek-Reasonix ships as a single static Go binary — a config-driven terminal AI agent that slashes token costs via DeepSeek's prefix cache and runs external tools as MCP subprocesses.

TencentDB-Agent-Memory is an open-source hub that distills conversations into L0-L3 layers, extracts Skills & knowledge maps. Eliminates cold-start & context repetition for multi-agent systems. Drop-in memory layer for any agent stack.

[GitHub Trending] TencentCloud/TencentDB-Agent-Memory

Four Signals — The Wire

foursignals.dev

Bhived's MCP gateway uses a five-stage loop to avoid shared agent memory amplifying mistakes. It fuses Qdrant, FalkorDB, and graph search, returning warnings and disputes. Write-backs link to query_id for failure tracking.

How we designed shared lessons for AI agents without trusting every write-back

Four Signals — The Wire

foursignals.dev

Uber open-sourced ADR: enterprise security for AI agents. Includes Sensor, ADR-Bench (300+ tasks, 133 MCP servers), and two-tier detector. Accepted to MLSys 2026. Evaluate open-source Sensor and Detector to monitor agent behavior.

[GitHub Trending] uber/ADR

Four Signals — The Wire

foursignals.dev

Better AI agents now refuse to proceed over label mismatches. METR's time horizon, FixedBench's 35-65% undesirable changes, and 400k sessions show rigid workflows backfire. Audit: replace work-generating constraints with boundary…

When Better Models Make Old Agent Workflows Worse

Four Signals — The Wire

foursignals.dev

Can your team change a delivery address safely? Boundary drift forces global context. Restore locality by redistributing mechanics and exposing essential policy.

Article: Enabling Evolutionary Architecture Through the Preservation of Change Locality

Four Signals — The Wire

foursignals.dev

HashiCorp ships vault-kube-kms beta: K8s clusters can now use Vault Enterprise as KMS provider for encryption at rest, separating key storage from data storage. Requires control plane manifest changes.

HashiCorp Ships Public Beta of Vault Kubernetes Key Management

Four Signals — The Wire

foursignals.dev

Monthly review agents drown in context debt. Four storage roles—working context, durable state, evidence, deliverable—plus structured JSON checkpoints preserve decisions, not summaries.

Long-Running AI Agents Accumulate Context Debt

Four Signals — The Wire

foursignals.dev

Parsing C's sizeof operator is surprisingly tricky. Syntactic ambiguities create edge cases that challenge compilers and static analysis tools. Understanding these is vital for reliable devtools.