A developer has published a two-pass method for catching AI-writing tells in technical drafts, pairing a small Python regex sketch that flags vague attribution, contrast frames, generic conclusions an…
A developer at Stepthrough, a free hands-on AI and software training service for white-collar workers, built a lesson showing how to fact-check AI-generated leadership decks made with Gamma. In the le…
A developer built a coffee-store support bot in n8n 1.123.84 running Qwen 1.8B via Ollama and tested it against 20 prompt injection payloads, each run three times. The naive workflow let 12 of 20 inje…
A developer published a set of Claude Code PreToolUse hooks that block destructive shell commands such as recursive rm on / or ~, git push --force, git reset --hard, and curl piped to sh by exiting wi…
A developer at Ziztu updated the company's AI YouTube Video Maker to constrain generated visuals, locking a single art style across a 3-to-60-minute faceless video, limiting props to those named in th…
A developer rebuilt a 2001 DSP-based robot-vision pipeline and pitted it against two local vision models, Gemma and Qwen, on synthetic frames deliberately corrupted with analog-TV ghost echoes. The cl…
A developer tested 21,812 MCP server tool lists against the Batch APIs of Claude Haiku 5.5, GPT-5-nano and Gemini 2.5 Flash, finding that each vendor validates tool lists before the model sees them an…
A developer fine-tuned Qwen2.5 1.5B Instruct on self-generated agent trajectories using free Kaggle T4 GPUs, with roughly 60% of the pipeline written by DeepSeek v4.1 Flash and the rest by GLM 5.3 Fla…
An autonomous AI agent writing as Feldspar for L3Digital LLC published a guide identifying three Supabase access-control bugs that single-user demos cannot surface: permissive RLS policies such as `us…
A developer published three permission presets for Claude Code's `.claude/settings.json` — strict, balanced and autonomous — that control which shell commands the agent may run without confirmation, a…
Sunny Chokshi, a PhD student at the University of the Cumberlands, benchmarked 16 open-weight LLM code reviewers on 180 Python snippets across nine cryptographic misuse classes, running 8,640 scored c…
Sunny Chokshi, a PhD student at the University of the Cumberlands, tested 16 open LLM code reviewers on 90 paired Python snippets covering nine classes of cryptographic API misuse, finding that writin…
Julien Wuthrich released Tracely, an MIT-licensed open-source project that turns failing production agent traces into deterministic CI regression tests by recording tool and model outputs as replayabl…
The open-source hmharness project (MIT) traced a multi-entry agent framework's inconsistent behavior to a "same-name, different-engine" split, where its BrowserOS assistant panel ran the full runAgent…
A developer running the random chat and gaming site Fyreflock built a server-side referee, GameSession, to let language models play two-player games through an MCP connector without leaking opponents'…
A solo developer building Refmade, an Electron desktop wrapper that drives the user's own installed Claude Code and Codex binaries via stream-json and JSON-RPC over stdio, found during end-to-end test…
HopBase has made Midjourney v8.2 (model ID midjourney-v8-2) available through its OpenAI-compatible image generation API, submitting jobs via POST /v1/images/generations and polling results through GE…
A developer proposes Capability-Oriented Development (COD), an approach that makes architectural relationships explicit so coding agents can find the owner of a shared capability, its supported option…
The developer behind Docustay, an AGPL-3.0 e-signature server with MIT-licensed SDKs, published a walkthrough showing how to send documents for signature programmatically, embed the signing form in an…
An API gateway operator serving both model families compared GPT-5.6 and Claude Sonnet 5 for coding work, finding Claude Sonnet 5 more consistent on multi-file changes in large repositories while GPT-…