Recently updated open-source projects

Claude Harness scaffolds a repository-level working framework for AI-agent work. It creates concise rules, active and completed execution plans, lessons, and a technical-debt tracker, then uses failing tests to lock behavior before implementation. Review, phase-finish, and lesson-promotion steps keep the context current over longer projects, making it useful for new codebases, refactors, and durable maintenance.

AgentsMesh provides a control layer for running terminal-based coding agents across a fleet of self-hosted runners. Each agent receives an isolated Git worktree, credentials, branch, and live terminal stream, while a shared console adds scheduling, autopilot, collaboration channels, and ticket tracking. Its BSL 1.1 license requires a commercial license for production use until the stated change date.

3-Layer Synthesis Kit preserves your notes and runs three generative passes: free ideation, forced combinations of atomic fragments, and a cross-pass between the two. Surviving ideas are grouped into four tiers by how strongly they connect to the source notes, with a Mermaid diagram in the output. It proposes possibilities rather than organizing notes or deciding which idea is valid.

TanStack Agent Skills gives coding agents structured rules for Query, Router, and Start. The guidance covers cache keys and invalidation, mutations, error handling, type-safe routes, loaders, search parameters, navigation, server functions, middleware, authentication, SSR, and security. Install the skills into a project with the cross-agent command, then compatible assistants can load them for relevant development or review tasks. The repository is a best-practice reference; it does not install or run the TanStack libraries or application.

Embedded Debugger MCP connects AI clients to probe-rs for probe discovery, target control, memory access, breakpoints, flashing, and RTT communication. The same package includes setup-oriented CLI commands and a skill for Codex and Claude Code. Hardware work requires a compatible debug probe, a supported ARM Cortex-M or RISC-V target, and correct SWD or JTAG wiring; destructive erase and write operations are gated by configuration.

ai-phrase-check uses a Detect, Suggest, Apply flow to spot recurring AI-style phrases in English and German and offer more human alternatives. The Claude Code skill and its in-progress Next.js demo share the same Markdown phrase lists, with user gates between stages; browser detection runs locally, while LLM suggestions use a user-supplied key stored only in session storage. The public demo is not live yet and there is no backend.

cc-statistics reads local records from Claude Code, Gemini CLI, Codex CLI, and Cursor to combine token use, estimated costs, sessions, tool calls, and project activity. Use its CLI or web dashboard to compare tools, search and resume sessions, generate periodic reports, or set usage alerts; macOS also has a native menu bar app. Core analysis needs no account or API key, while the Windows tray build remains a development preview.

These Datadog skills let compatible agents manage monitors, search logs and traces, analyze agent-observability experiments, investigate CI failures and flaky tests, and work through audit-trail incidents. Individual workflows use dd-pup, Datadog MCP toolsets, or REST APIs, so the required account access, permissions, and API credentials vary by skill.

video-mockup-extraction uses ffmpeg to turn a screen recording into a three-level visual reference: contact sheets for overview, 1280px frames for selection, and full-resolution stills for close inspection. It combines scene detection with interval sampling to cover both hard cuts and scrolling, and works as either a Claude Code skill or standalone CLI.

fullstack-ts-scaffold regenerates a secret-free full-stack TypeScript starter with a Next.js 15 client, Express API, rotating JWT refresh tokens, a worker, PWA, OpenAPI, tests, and CI. It supports Postgres/Prisma, MongoDB/Mongoose, and Supabase backends from one template. Full verification uses Docker-backed databases; --fast limits the gate to lint, type-check, and build.

fullstack-ts-scaffold regenerates a secret-free full-stack TypeScript starter with a Next.js 15 client, Express API, rotating JWT refresh tokens, a worker, PWA, OpenAPI, tests, and CI. It supports Postgres/Prisma, MongoDB/Mongoose, and Supabase backends from one template. Full verification uses Docker-backed databases; --fast limits the gate to lint, type-check, and build.

Codex Infinity extends the OpenAI Codex CLI with automatic next steps, testing, new improvement ideas, and chained goals. It can continue after each response or start another goal when one finishes, while detecting model providers from available credentials. Use it when unattended continuation is intentional, since the agent can keep making changes.

SWE-smith creates Docker execution environments from GitHub repositories, synthesizes tasks such as file localization and program repair, filters instances that break one or more tests, and generates issue text. The resulting tasks and trajectories can support SWE-agent training or evaluation. It requires Docker and was developed for Ubuntu 22.04; Windows and macOS are not planned targets.

Give Claude a topic, outline, notes, document, or previous PowerPoint and Presentation Forge writes the slides, builds a theme, and bundles the engine, fonts, and images into one offline index.html. Decks include presenter mode, notes, a timer, progressive reveals, and reusable brand styles. Saving edits back from the browser works only for locally opened files in Chromium-based browsers, and building requires Python 3.10 or later.

claude-brain installs an interconnected Markdown knowledge base in a project, with session notes, a dense index, an append-only log, and a source registry. Its ingest command captures the current conversation, while query uses the wiki as a source for answers. It works with an Obsidian-style setup and can optionally use Notion as a source.

Claude Engineer provides CLI and web interfaces where Claude can identify a missing capability, generate the corresponding tool code, load it dynamically, and keep using it during development work. It also manages persistent conversations, context and token usage, tool chaining, and debugging output, with optional Python execution through an E2B sandbox. Its core workflow depends on Anthropic Claude 3.5, and some tools require additional service credentials.

aham-ppt turns documents, data, and notes into professional, editable PPTX decks through a staged process: load brand rules, extract claims, build the narrative, approve sample pages, design the full deck, and run a delivery check. It offers 39 parameterized layout IDs, three visual styles, and chart or annotation APIs. The skill favors one conclusion per page and evidence-backed wording over quick template output.

AgencyBench V2 covers six capabilities across game, frontend, backend, code, research, and MCP work, with 32 real-world long-horizon scenarios and 138 tasks. A user-simulation agent supplies iterative feedback while a Docker sandbox runs functional and visual checks using rule-based, vision-based, and LLM judges. Scenarios require model and evaluator configuration, and game or frontend cases also need the remote Docker sandbox.