标签：#ens

I spent two months building Vestibule, an open-source Python framework for the boring layer of RAG ingestion — stable document IDs, a state ledger, error classification, per-vertical governance. The parts every team struggles with once the demo works and production doesn't. Most of the code wasn't typed by me. Four AI agents did the work — one wrote designs, one reviewed them, one implemented, one reviewed the code — all through real GitHub pull requests, with me signing off at every gate. The result: twelve components, three releases, 878 tests. Two moments defined the whole experience. When the process caught what I couldn't The trickiest component provisions vector indexes on first use, safely even when workers race each other. Its design was rejected and revised five times before any code existed. In the first round, the reviewer agent found a genuine race condition: a worker still inside a slow index-creation call (~390 seconds with retries) would look stale (the threshold defaulted to 300 seconds), lose its claim to a waiting worker, and now two workers create the same index. A production race, in the default configuration, spotted by one AI reading another AI's design — before a single line was written. When green tests lied to me After v0.2 shipped, I wrote a quickstart script and ran the pipeline the way a stranger would — for the first time. pip install didn't work. At all. A packaging conflict made the whole framework uninstallable, while 483 tests sat green. An hour of actually using it turned up two more: a default model name that had never once worked against the real SDK, and an import that took down an entire package when an optional dependency was absent. What went wrong wasn't the tests — it was what they measured. They proved the code agreed with itself: same working tree, same mocked seams. Nothing ever checked the world a user lives in: clean machine, real install, real SDK. Passing tests and a working product turn out to be two different claims

skillcheck is a static analyzer for SKILL.md files, the format agents like Claude Code, Copilot, Codex, and Cursor use to load reusable skills. It validates frontmatter, scores description discoverability, checks file references, enforces token budgets, and flags cross-agent compatibility issues. No network calls, no LLM calls, no file mutations. Runs as a CLI, a GitHub Action, or a pre-commit hook. pip install skillcheck skillcheck skills/ Latest pass was hardening and accuracy, not features. Here's what changed and why. Description scores went up. Skills that were scoring low because the scorer was broken will now see a jump in scoring. Median across the reference corpus went from 75 to 90. --explain-score also now tells you which pattern hits or misses instead of just a number. The score exists to predict whether an agent will actually find and trigger your skill, so a scorer that under-credits good descriptions defeats the point. The fix was validated against real-world skills, and the separation held: filler still scores 28-65, well-written descriptions 85-100. Corrupt files now fail cleanly instead of crashing. Before, a bad history ledger or non-UTF-8 skillcheck.toml above the skill dumped a Python traceback. It's now a clear error naming the file and byte offset (exit code 2). Config discovery walks up the directory tree, so one bad file could break every scan under it. Now every untrusted read (ingest, history, config) goes through the same guard before parsing, so they all reject the same way. README has been corrected in regards to token estimates. Without tiktoken, expect roughly 20-30% over-estimation, so install the extra if you're near a budget limit. The offline heuristic feeds the budget checks and its accuracy had never actually been measured, just assumed. It's benchmarked against tiktoken across the full corpus now, and the documented numbers are the measured ones. pip install "skillcheck[tiktoken]" The rest of the pass is invisible on purpose: f

mcp-drift-monitor detecta cambios no autorizados en servidores MCP (Model Context Protocol). Implementa el control primario faltante descrito en arXiv:2608.00997 : un barrido completo periódico del catálogo que re-descarga todos los servidores y recomputa hashes. Problema arXiv:2608.00997 ( MCP Registry Drift: A 88.6-Day Measurement of 19,099 Servers ) reporta un punto ciego crítico: los enfoques tradicionales de detección de cambios fallan en identificar dos modos de fallo: Cambios silenciosos — un servidor cuyo hash de descripción cambia, pero el monitor ya lo conocía y lo rankinga por historial pasado. Nuevas adiciones — servidores que aparecen en el registro sin que el monitor tenga registro previo. El paper mide 15,845 eventos de cambio, 19,877 adiciones y 911 eliminaciones, pero los modelos que rankean por historial previo pierden una fracción significativa de estos eventos. Este monitor cierra esa brecha con el control primario que el paper propone pero no implementa: un full-catalog sweep periódico. Solución mcp-drift-monitor implementa un motor de diferencias único ( compute_events ) que sirve tanto para polling incremental como para barridos completos. No hay lógica duplicada. Cada vez que un hash de descripción cambia, el motor revalida el contenido ( len(drifts) > 0 es el único disparador). Si el registro responde 429, aplica backoff con Retry-After . Si el payload está malformado, lanza SchemaDriftError y registra el payload ofensor a nivel ERROR. Arquitectura core/ diff.py — CatalogEntry, DriftEvent, NewArrivalEvent, RemovalEvent, compute_events hasher.py — normalize_description (NFC), hash_description state.py — StateStore (sqlite), FetchStatus, removed flag, get_all_hashes poller.py — Poller.fetch_catalog, PollConfig, SchemaDriftError, backoff sweep.py — run_sweep (control primario), SweepReport calibrate.py — replay (FR-6), ReplayReport, external validity vs panel Resultados de calibración El monitor se calibró y verificó contra el panel real del pa

I maintain a small comic and manga viewer component for React called react-comic-viewer . The other day I was poking around npm and noticed something odd — there were three other packages with basically my package's name, published by other people. All three were forks of mine. Same description, same repository URL pointing back at my repo. None of the three authors had ever opened an issue or a pull request on my side. What the fork changed The oldest fork was made about three months after I first published, and it kept going for almost a year. Its version number ran ahead of mine at the time — I was on 0.3.5 while the fork was on 0.6.3. So I read the diff. Honestly, it was more useful than any issue would have been. The commit messages alone told the whole story: remove sass fix: support className props use Hotkeys and a new example file called controlled.tsx The sass one I'd already fixed. The className one I'd fixed too, about a year later. The controlled.tsx one I had never fixed. Not in four years. The part I never fixed Here's what their example looked like: < ComicViewer currentPage = { currentPage } isExpansion = { false } onTryMoveNextPage = { ( nextPage ) => { /* ... */ } } onChangedCurrentPage = { ( page ) => setCurrentPage ( page ) } pages = { pages } /> And here's what my component actually accepted: < ComicViewer initialCurrentPage = { 0 } initialIsExpansion = { false } onChangeCurrentPage = { ( page ) => { /* ... */ } } pages = { pages } /> The initial prefix is the whole problem. My component would take a starting page from you, and then never let you touch it again. It owned that state for the rest of its life. That's fine for a demo. It's pretty bad for anything real. You can't jump to a page from a table of contents. Syncing the current page with the URL doesn't work either. And if a chapter needs to be purchased first, there's no way to step in and stop the move. Every one of those needs the parent to be in charge, and the parent never was. Maki

Every coding agent says it is efficient. Almost none of them publish the bill. So we ran the boring experiment: the same bugfix, the same model, the same API, the same prices, and a byte identical prompt, once through opencode and once through the coding agent inside Locally Uncensored. Headline: opencode averaged 2157 credits over three runs. Our 2.6.6 agent finished the identical task for 1298 . That is about 40 percent less, and even the cheapest opencode run came in 29 percent above our number. The interesting part is not the headline. It is why the gap exists, and it is not the reason most people guess. Setup A cost comparison is only worth reading if everything that drives cost is nailed down. What was held constant: Held constant Value Task Fix a failing test in a small npm repo, then commit Repository Three files, a one line bug in add.js , tests red at the start Prompt Byte identical, sha256 29cec6c3...cf62687 Model deepseek-ai/DeepSeek-V3.2 Endpoint The same OpenAI compatible API for both agents Prices Same account, same tier, same per token rate Counting One wire proxy in front of the API, credits read before and after every run opencode 1.18.21 from npm, wired as an OpenAI compatible provider, opencode run --auto , otherwise defaults Success was defined before the runs, not after: npm test passes exactly one commit, with the required message only add.js changed clean working tree at the end All four runs cleared that bar. Nothing failed, so cost is the only variable that moved. The numbers Run Credits Requests Prompt tokens Success opencode, run 1 1679 8 98,789 yes opencode, run 2 2433 11 146,058 yes opencode, run 3 2358 11 146,387 yes Locally Uncensored 2.6.6 1298 16 74,629 yes Locally Uncensored 2.6.5 4395 30 257,270 yes Read the last row first. Our own shipped agent from one release earlier is the most expensive thing in that table, by a lot. This is not a chart built so that we win by construction. It is a chart that shows what one efficiency pass is

Hello I'm creating CrowdGPT , an open-source project which allows training of a LLM (Large Language Model) in a decentralized way, where each user contributes to making the AI better with whatever data they want. The idea is simple: instead of one machine owning the entire training run, let many people contribute small training jobs and periodically merge those updates into a shared model. The system is based on a centralized server (lightweight) that receives every client training, then "merge them back" to the main model. This system prevents threats or malicious updates by doing cross-client verifications (provides a proof of work). The users that train the model are being put on a leaderboard, rewarding their contribution. Data is taken from a curated dataset on Hugging Face (which means no personal data is ever used during training). However, users can push new text to this dataset (which is then moderated and validated). If you're curious, here is the GitHub: https://github.com/Vxtzq/CrowdGPT Here is the website: https://www.crowdgpt.net The best way to help me is to either: Give feedback on what must be changed to make it a fully finished project. I'm mainly looking for criticism: what would stop you from running this on your own GPU? Contribute to the project by becoming a part of the network (coming soon) Star the repo on GitHub ⭐ It helps a lot :)

Pub.dev Package: Link GitHub Repository: Link Imagine getting into a serious car crash in a remote...

Once you use Claude Code or Codex for real work, a total usage number stops being enough. You want to know which change consumed it. I did not build agent-cost because I had missed the existing token and cost trackers. I knew about multi-agent reporting CLIs, local dashboards, and OpenTelemetry-style observability stacks. I had even built a similar view in Notion before. The problem appeared when I tried to use that kind of reporting in an operational workflow. I needed agent logs to stay on the machine. I wanted a small runtime dependency surface, custom metrics I could audit, and a machine-readable result that another tool could consume. Most importantly, I needed session measurement and task attribution to remain two different claims. I did not need another universal dashboard. I needed a boundary underneath the dashboard that could answer: is this number supported well enough to enter task accounting? A measurement layer below the UI Different tools optimize for different jobs. A broad CLI such as ccusage is useful when coverage across agents matters. Local interfaces such as token-tracker or AgentMeter are a better fit for visual exploration of projects, sessions, subagents, and tools. An OpenTelemetry stack is the natural choice for fleet-level metrics, logs, and traces. Those are not inferior versions of agent-cost . They serve different use cases and trust models. The layer I wanted looked like this: local observations -> auditable normalized facts -> explicit pricing status -> caller-selected sessions -> task-attribution policy -> optional dashboard / Notion / spec-lane agent-cost reads logs that Claude Code and Codex CLI have already written locally. It normalizes each usage event into a fact with a model, token kind, timestamp, and count. At runtime it makes no network calls and declares no Python runtime dependencies. Its price catalog has a version and SHA-256 digest, both carried into machine-readable output. That “zero-network” claim is deliberately l

GitHub热门项目 | Vane is an AI-powered answering engine. | Stars: 36,373 | 46 stars today | 语言: TypeScript

GitHub热门项目 | Database manager for MySQL, PostgreSQL, SQL Server, MongoDB, SQLite and others. Runs under Windows, Linux, Mac or as web application | Stars: 7,269 | 6 stars today | 语言: JavaScript

GitHub热门项目 | Community plugin marketplace for Claude Cowork and Claude Code. Read-only mirror — submit plugins at clau.de/plugin-directory-submission. | Stars: 547 | 141 stars today | 语言: Python

GitHub热门项目 | TSP自托管、零运维的 A 股「选股 + 监控 + 回测」量化工作台 | 基于 TickFlow 数据源 | LLM能力驱使策略定制+个股分析+复盘 | 自由接入第三方数据源与个性化扩展数据 | 个人开源 ,非TickFlow官方项目 | Stars: 3,448 | 90 stars today | 语言: Python

GitHub热门项目 | Sub2API 一站式开源中转服务，让 Claude、Openai 、Gemini、Grok订阅统一接入，支持拼车共享，更高效分摊成本，原生工具无缝使用。 | Stars: 38,723 | 264 stars today | 语言: Go

Kintara is a self-hosted document library and reader that runs in Docker and watches a folder you already have. Drop PDFs, Markdown, or text files into the directory and it indexes them automatically, extracts searchable text and metadata, generates thumbnails, and makes the whole library available through a browser or installable PWA. It has libraries, collections, tags, full-text search, highlights, favorites, reading progress, private library sharing, and GitHub OAuth. I have been working on Kintara for a few months, and the architecture actually changed pretty dramatically while I was building it. Kintara originally had a Tauri desktop shell, but I eventually realized that isn't what I wanted at all. So I ripped the desktop layer out and rebuilt it around one Rust server that serves both the API and frontend. Now I can point Kintara at a NAS folder and open the same library from my desktop, laptop, tablet, or phone. The thing I really love about this app is the optional AI features. I added an option to use OpenAI or Gemini, and with so few tokens being spent, it's a fraction of a cent to use most of them, aside from the cover image generation, which is bit more, but makes the library look so much prettier! 😄 Anyway, I wanted AI to be a tool inside the library rather than taking the thing over, and I wanted it to be fully optional, so if you're one of those "Ew, AI is in this app" people, you just don't turn it on and it's like it doesn't exist. What the AI can do is summarize documents, suggest metadata and fill in those blank spaces, generate cover images for docs that don't have a cover, search the library for docs, or you can just chat with it about your docs. Find is a pretty great AI feature I think. Instead of letting the model vaguely tell you that something appears "somewhere in the document," Kintara asks for actual passages with page numbers, verifies the quote against extracted page text on the server, then verifies it again against the rendered PDF.

GitHub热门项目 | VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages. | Stars: 11,154 | 125 stars today | 语言: Python

At the first step, I picked up the SurrealDB project for contribution. I didn't know how I could help this project become better. So I asked my beautiful OpenCode to find parts of the project that could be better. It suggested this file of the project(core/src/val/value/get.rs) to me and said it has a double-cloning issue. So I opened up VS Code, and I started checking the issue. The code was something like this: let mut a = Vec :: new (); for v in v .iter () { let cur = v .clone () .into (); if stk .run (| stk | w .compute ( stk , ctx , opt , Some ( & cur ))) .await .catch_return () ? .is_truthy () { a .push ( v .clone ()); } } First Optimization: As you can see at line 3 and line 9, we have multiple clones from a single document. I thought about how I could fix this issue; I went to see the CursorDoc structure because the first clone is converted to it: #[derive(Clone, Debug)] pub ( crate ) struct CursorDoc { pub ( crate ) rid : Option < Arc < RecordId >> , pub ( crate ) ir : Option < Arc < IteratorRecord >> , pub ( crate ) doc : CursorRecord , pub ( crate ) fields_computed : bool , } impl From < Value > for CursorDoc { fn from ( val : Value ) -> Self { Self { rid : None , ir : None , doc : val .into (), fields_computed : false , } } } #[derive(Clone, Debug)] pub ( crate ) struct CursorRecord { /// The underlying record, shared via Arc for copy-on-write record : Arc < Record > , } impl CursorRecord { // .... // /// cloning. Otherwise the value is cloned. pub ( crate ) fn into_owned ( self ) -> Value { match Arc :: try_unwrap ( self .record ) { Ok ( record ) => record .data , Err ( arc ) => arc .data .clone (), } } // .... // } impl From < Value > for CursorRecord { fn from ( value : Value ) -> Self { Self { record : Arc :: new ( Record :: new ( value )), } } } I saw that the value passed through CursorDoc is directly stored in a field in CursorRecord without any changes, and it is accessible using .into_owned() from CursorRecord. That is the solution; I edited the

Every open-source tool has the same brutal first 90 days: zero users, zero signal, no idea whether anything works. I have shipped several free CLI tools and browser tool sets. This is the exact funnel I use — no ads, no paid growth, no "build in public" theater. Just a repeating sequence of small, concrete actions. Step 1: Make the Tool Trivial to Try The first rule: npx must work. If a reader has to install, configure, and read a README before running the first command, the funnel is already broken. npx @wuchunjie/dotguard . That is the entire onboarding. Zero dependencies, no config, instant output. The first 10 seconds decide whether the reader comes back. Step 2: Publish One Article Per Angle Not one article. One per angle , spread over time: Tutorial — "Scan your .env files in 1 command" (the how) Comparison — "Why I stopped using X" (the why) Listicle — "5 tools for Y" (the discovery) Workflow — "My dev setup" (the context) Security/devops — "Your CI is missing this" (the fear) Each article targets a different search intent. A developer looking for "pre-commit secret scan" lands on article 4, not article 1. The funnel is wide because the angles are wide. Step 3: Cross-Link Everything Every article mentions every tool. The footer of a snippet article lists the scaffolder and the scanner. The GitHub repo links to the articles. The npm README links to the articles. The effect is compounding: a reader of article 3 meets four tools, not one. Your content becomes a network instead of a pile. Step 4: Make the GitHub Repo the Hub The repo README is the landing page that never goes stale: One-line description per tool Install/run commands (copy-paste ready) Links to every article A donation link, present but quiet GitHub is where developers actually trust. Stars and forks are the signal that converts "interesting article" into "let me try it". Step 5: Add the Quiet CTA One line at the end of every article: If this saved you time, a Ko-fi keeps the next tool coming. No

When I started GSoC in May, my plan was to build a runtime sandbox for community plugins. By week two my mentor had talked me out of it, and I ended up spending the rest of the summer building a review queue instead. This post is about how that happened and what I actually shipped. Quick summary Project: Community Driven Plugin Ecosystem for OWTF Org: OWASP Foundation Mentors: Abraham Aranguren, Viyat Bhalodia What got shipped: Six pull requests against owtf/owtf , around 6,000 lines of Python and TypeScript, 153 backend unit tests, and a trust model doc. Working mirror of this post: gist If you only want the code, here are all my PRs on OWTF . The problem I was trying to solve OWTF is a security testing framework, and until this summer its plugin catalogue was static. If you wrote a detection for some new attack pattern, your options were: open a PR against the framework itself (high bar, slow), or keep the plugin to yourself. Most useful plugins never made it upstream because of that. The Community Plugin Marketplace fixes this. Any authenticated user can upload a Python plugin through the web UI. The plugin is validated at upload time, lands in a pending queue, and waits for an admin to look at the source. Once approved, the plugin gets mirrored into OWTF's standard plugin table. From that point on, the runner, the worklist, and the report generator all treat it exactly like a built-in plugin. The pivot My accepted proposal called for a sandbox. Community plugins would run inside something like a subprocess with dropped privileges, so that a malicious plugin could not do too much damage. Then Viyat said this in Slack: A sandbox in Python that talks to the same postgres, the same file system, the same target scope as OWTF itself is not really a security boundary. I sat with that for a couple of days and realised he was right. A plugin that runs inside OWTF has to see the target, has to read config, has to write results. Any "sandbox" I put around that is going to

Tu agente de código lee tu workspace. Un archivo envenenado en cualquier rincón puede llevar instrucciones que el agente ejecuta. ¿Sabes qué fracción de tu workspace tiene que leer para que eso ocurra? topowatch mide eso. El problema no es el prompt, es la topología El paper Workspace Topology as an Attack Vector in Agentic Coding Assistants (arXiv:2608.14876, Day et al., 2026) demostró algo que intuíamos pero no medíamos: la topología del workspace afecta mediblemente el Attack Success Rate (ASR) de la inyección indirecta. Los entornos altamente modulares muestran ASR significativamente menor que los planos. La razón es mecánica: si el agente acota su lectura al módulo de la tarea, nunca llega al archivo envenenado. Si hace un wide read de todo el workspace, lo lee siempre. Qué es topowatch topowatch es una herramienta de línea de comandos que, dado un workspace, mide el ASR de una inyección indirecta de referencia bajo varias configuraciones de topología, y reporta qué estructura minimiza el ASR. Fundamentado en arXiv:2608.14876. Determinista y reproducible sin claves ni red: usa un agente sintético configurable y un fixture con tres topologías (monolito, modular, nesting profundo). pip install -e ".[test]" topowatch --json Resultados Sobre el fixture de referencia (200 trials, semilla fija): Topología ASR % leído Monolito (plano) 1.000 100% Modular (acotado) 0.000 28.5% Nesting profundo 0.000 66.6% El reporte incluye read_budget (fracción del workspace que lee el agente) y el veredicto del defense contract: modular < monolito . Honestidad sobre v0.1 v0.1 usa un agente sintético , no un coding assistant real (Claude Code / Codex). El claim "modularidad → ASR menor" está anclado al fixture reproducible, no a una medición contra un assistant real — eso es v0.2 (feature 002). El objetivo de v0.1 es darte una herramienta para medir y recomendar modularidad, no simular un ataque completo. Roadmap v0.2 : medición contra coding assistants reales (sandbox, sin credenciale

For three years, the biggest complaint about Mojo was not the syntax, the performance claims, or the missing ecosystem. It was that the compiler was closed. You could read the standard library, you could file issues, but the thing that turned your code into GPU machine instructions was a binary you had to download on faith. For a language whose creator, Chris Lattner, built his reputation on LLVM and Swift, two of the most open projects in compiler history, that sat badly with a lot of developers. Then came the strangest possible sequence. Qualcomm announced an all-stock acquisition of Modular on June 24, 2026, valued around $3.92 billion at announcement. The deal closed at the end of July. Mojo hit version 1.0 the following week. And on August 18 at ModCon, Modular open sourced the entire compiler and toolchain under Apache 2.0 with LLVM exceptions. A chip company bought the language, and only then did the source drop. The Hacker News thread reached 409 points, and the reaction splits into two camps that basically never overlap: people who say "finally, I can try this," and people who say "too late, the window closed." Both are worth listening to, because the honest answer to whether Mojo matters now depends on what you actually do with Python. What Actually Got Released The whole toolchain, not a teaser. The modular repository on GitHub now contains the Mojo compiler, the tooling, and everything needed to build the language from source. One command builds the compiler and runs a Mojo file against it: ./bazelw run --config = build-mojo KGEN:mojo -- run hello.mojo That is a real bar to clear. This is not "source available with a look-but-do-not-touch license." Apache 2.0 is the same license family as the rest of the LLVM world, and the LLVM exceptions expand what you can do with distributed binaries. You can fork it today if you want. But not contributions, yet. The announcement is explicit: Modular is not accepting contributions to the compiler and tooling right no