Recently updated open-source projects

SWE-Gym provides 2,400 real software-engineering tasks drawn from 11 Python repositories, together with repository context, executable environments, and test-based verification. Researchers can use it to generate trajectories and train agents or solution verifiers. Reproducing the experiments also involves the published dataset, prebuilt Docker images, and instructions for OpenHands or MoatlessTools.

Mirobody has a C++ core that runs inside Android and iOS apps, in a desktop wrapper, or as a self-hosted server. It brings wearable, laboratory, and clinical records behind a common interface, with FHIR endpoints, long-term memory, MCP tools, and several LLM providers. Health data can remain on the phone or the operator’s own infrastructure, but enabling an external model, synchronization, or vendor integration can send selected information outside that boundary. The architecture supplies privacy and compliance controls; the deployer remains responsible for the final HIPAA or GDPR posture.

aham-word centralizes page, font, color, paragraph, table, header, footer, and logo values in word-tokens.json, while a specification covering 16 document elements guides a Python generation engine. It produces editable DOCX files with covers, automatic tables of contents, heading levels, horizontal-rule tables, and page-number fields. Changing the token file rethemes the output without code edits, and missing facts are marked for completion rather than invented.

HeiGe-SuanMing uses scripts to calculate Bazi, Zi Wei Dou Shu, Meihua, Liuyao, and Qimen charts before applying documented reasoning steps. It can produce a visual HTML reading with traceable grounds and separate options for compatibility, timing, or focused analysis. Interpretations are framed as reference, not medical or life certainty.

angular-material-skill generates an AI reference skill for Angular Material and CDK from the official angular/components repository. A manifest pins the source tag, while release checks and a manual sync regenerate only changed references for review. The project is still marked as a v1 work in progress, and generated updates are never committed automatically.

CodeMachine is an orchestration layer for AI coding CLIs such as Claude Code, Codex, and Cursor. Define a workflow once and it handles step execution, context passing, and agent coordination. Workflows can range from interactive to fully autonomous, divide work among multiple agents, run tasks in parallel, and persist for hours or days. The underlying coding CLIs must be available separately.

claude-design-idea-to-ready defines design tokens and a CLAUDE.md rules block before screens are produced. It then builds components, states, interactions, and accessibility checks in order, generates the UI, and verifies the result against those rules, making it useful for projects that need consistent branded screens.

generate-walkthrough investigates entry points, journeys, data models, parameters, and boundaries, then writes a self-contained HTML walkthrough whose claims carry file:line evidence. A schema-pinned JSON sidecar can feed the companion extractor for OpenAPI 3.0.3, Postman, and AWS-call artifacts, with unrecoverable details marked as coverage gaps. It is experimental and works best on small to medium codebases or a scoped monorepo slice.

AgentPack performs a local preflight before an agent edits code. Given a task, it ranks likely files, tests, rules, skills, commands, and warnings, then renders a compact context pack with reasons and receipts for what was included or skipped. It uses local analysis and reusable caches rather than cloud indexing or model calls, and its benchmark measures file-selection overlap—not guaranteed task success—so source inspection and tests still matter.

kb-with-receipts turns scattered product documents, design files, and tutorial videos into a Markdown knowledge base with file-and-section citations, then feeds it to a no-code Q&A tool and a reusable agent skill. Its method preserves numbers and rules verbatim, tags every section, lets newer material supersede older versions with a note, and runs question-based QA plus integrity checks. It is tool-agnostic, but the source material still needs domain experts to catch drafts or unbuilt ideas.

cowork-to-code-bridge connects Claude Cowork or a browser chat to Claude Code on a macOS, Linux, or WSL2 machine. It forwards approved tasks or scripts without opening network ports or using sudo, streams results back, and caches idempotent retries so work is not run twice.

Blindspot Audit gives an agent a repeatable way to find unknown unknowns, hidden risks, stale assumptions, and missing decisions across software, games, writing, research, content, and business plans. It records ranked findings, covered areas, and deferred checks in a durable ledger, so later audits focus on changes instead of repeating the same list. It supports Claude Code, Codex, OpenCode, Claude desktop, and plain chat.

Critic Gauntlet runs parallel adversarial critiques from model families such as Claude, Codex, Grok, and Gemini, then synthesizes major gaps, steel-manned alternatives, assumptions, six-month consequences, and a recommendation. It is aimed at expensive-to-reverse architecture choices, not routine bug fixes or refactors. Extra critics require their CLIs, APIs, and credentials.

CLITrigger combines project docs, planning, live terminals, autonomous tasks, and Git review in one command center. Claude, Antigravity, or Codex jobs can run in parallel or on schedules, each inside an isolated worktree, while completed diffs flow into a central accept, reject, or merge queue. It is available as a desktop app or a Node-based service for developers managing many AI coding tasks.

Distill mode researches a public figure's own words, documented decisions, outside accounts, failures, and criticism to create a sourced profile of recurring principles and where they break down. Apply mode uses that profile as one lens on a concrete decision and states what the framework misses. Nine profiles are included; creating new ones requires research access, and the result is not presented as the person's private view or a final verdict.

Agent Skills CLI manages skills across tools such as Claude Code, Cursor, Codex, and other coding assistants from one command line. It can install from a marketplace, public or private Git repositories, and npm packages, then update or synchronize selected targets. Lockfile-based installs, conflict checks, quality tests, and sandbox previews help users inspect collections before wider rollout. The CLI requires Node.js 18 or later.

This repository is a Claude Code marketplace for Glitchwerks projects. Its current listing is claude-wayfinder, which helps select agents and skills through typed, auditable scoring; a prospector plugin is listed as upcoming rather than available. Marketplace entries pin source commits so installs are reproducible, making the repo useful for users who want a small, version-pinned plugin catalog.

ai-title-generator turns an article or topic into Chinese titles for WeChat, Zhihu, and Xiaohongshu. Its professional tier maps each platform to one of three tech-media styles with a main and alternate title, while the Xiaohongshu tier uses first-person framing, emojis, and a pain-point hook. It is aimed at Chinese creators, product managers, and AI practitioners and avoids inventing hooks.