AWS launches Emulator for agents to test code against local AWS
Amazon says the tool is meant to stand in for more than 200 services that are too numerous to mock by hand.
Stories per day · 30 days
17 RSS
Relatedclaude-code12benchmarks11codex11developer-tools10prompt-injection7swe-bench7qwen6reinforcement-learning6benchmark5claude5openai5llm-agents4opencode4open-source4open-weights4software-engineering4supply-chain43d-generation3
197 stories, newest first · one card per story
Stories tagged #coding-agents
Klad paint app lets a coding agent drive it
The source-available project drew 24 points and nine comments on Hacker News.
Grok Build share links exposed apps, including SpaceXSI data
A passive DNS search found 43,601 grok.me links, and more than 35,000 opened stored app data.
1
Atuin Threads syncs coding-agent sessions across machines
From Atuin 18.24, opt-in end-to-end encrypted history can resume or fork sessions across Claude Code, Codex, Pi, Omp, and OpenCode.
1
OpenAI workshop covers Codex cloud handoff, voice mode and dots
Engineers said Codex Cloud version 2, voice control and personal dots still feel awkward for many developers.
Claude Code can commit secrets into public GitHub repos
Apiiro counted 2,539 credentials in nearly 2,000 public repositories, including production tokens and a medical database password.
Google Chrome repo steers coding agents toward modern web primitives
The Show HN post links GoogleChrome/modern-web-guidance and had 25 points and 4 comments.
Essay follows an engineer automating work with Claude
The piece is a narrative about sleep loss and escalating automation, not an incident report.
Claude Fable 5.2 coding edits follow prompt ownership wording
Calling the codebase the model’s led it to merge repeated code in about half of small-task tries, versus under a tenth otherwise.
Simon Willison suggests hiding AI PR descriptions behind details summary
AI agents like Claude Code generate overly detailed PR descriptions that read like slop, prompting the use of HTML details tags to hide them.
Stanford CS146S teaches AI-native engineering via open-source maintainer reviews
Podcast with Monaco's AI lead discusses Stanford course where students review open-source code for AI agents.
1
Embodied First Exam tests coding agents on 1,156 robot tasks
GPT-6 Astra in Codex, strongest of eight agents, succeeds on 34% of 456 core tasks.
Coding agents copy existing vulnerabilities into new code
Insecure neighbors cut secure backend implementations 3.4% and widen the C/C++ context gap to 35.4%.
ArmedAgent finds safety risks in coding agents that control robot arms
The benchmark covers 36 tasks in five scenarios, including indirect prompt injection that can move from digital resources into physical actions.
Frontier coding LLMs reward-hack unseen repository tasks
With answer leaks blocked, models still mined git history, the object store, and the filesystem.
Frontier3DBench benchmarks coding agents for 3D creation
GPT-6.1-sol leads object and scene levels on Frontier3DBench while reference reconstruction remains a key bottleneck
Video-Coder trains code-driven agents for video generation
Video-Coder achieves 7-point gains on PhyGenBench and 24-point gains on VBench-2.0 via agentic reinforcement learning
ReproBridge lifts ML reproduction success from 76.4% to 92.6%
Across 1,029 tasks from 387 experiments, refined artifacts also cut mean API spending by 19.7%.
Terminal-Bench 2.1 repairs improve coding agent benchmark accuracy
Benchmark maintenance debt leads to issues in 30% of tasks with accuracy gains over 10 percentage points after repairs.
Meta acquires AI coding startup Conductor
Conductor Cloud shuts down on November 20, and the desktop app stops receiving new models in early 2027.
Agent-Native Primitives improve real-world task resolution for coding agents
HEELO framework using eight primitive families raises Terminal-Bench 4.0 resolution from 37.3% to 55.5% and reduces task time on heelo.dev
arXiv study shows coding agents can compress LLMs
29 of 36 agent submissions pass verification, achieving 77.33% downstream accuracy on compressed models.
VeriHarm-Bench measures harmful success in LLM coding agents
Frontier agents achieve 95% task accuracy but only 37% safe success on VeriHarm-Bench.
Stop counting tokens start counting what ships
AI coding agents should be measured by cost per merged pull request rather than tokens consumed to track value delivered.
1