Recently updated open-source projects

mimic helps an agent turn MediaPipe Unity Plugin samples into real-time webcam interactions for pose, hands, face, gestures, objects, or segmentation. It covers intake questions, scene selection, landmark and coordinate references, 2D/3D mapping, and the thread-safe callback-to-Update integration pattern.

Claude Autopilot for Obsidian makes Claude read relevant notes before editing code, update the vault alongside the change, archive sessions, cross-link related work, and optionally sync through git. The pure-Python toolkit runs on macOS, Linux, and Windows; archives stay local and are secret-redacted by default, while pushing them requires explicit opt-in.

InferSpec reconstructs OpenSpec documents for brownfield systems by reading code, git history, project docs, and optional Jira, Confluence, or web sources. It creates one spec per capability, cites requirements to file lines or ticket IDs, marks ambiguity with GAP or TBD, and offers bulk scans, focused capability passes, and installed-skill updates without a separate model API key.

Win-C-Cleaner guides Windows 10 and 11 C-drive cleanup through eight stages, from temporary files and hibernation data to component storage, duplicate drivers, OEM software, and recovery packages. It requires an administrator PowerShell, validates deletion paths, defaults the first stage to dry-run, and adds explicit confirmations or backups to riskier operations. Some component-store cleanup cannot be rolled back, so the staged workflow matters.

Ask Claude, Cursor, Windsurf, ChatGPT, n8n, or any HTTP client whether a domain is available across more than 50 TLDs and where registration is cheapest. DigMyName verifies results through several sources and returns an unknown state instead of treating uncertainty as a taken domain. The public endpoint requires no API key, but per-minute and daily fair-use limits apply.

council-review sends a project, system, document, or plan to three to six blind reviewer lenses in parallel, then has a chair synthesize agreement, tensions, strengths, and prioritized recommendations. It is report-only and never edits the subject under review.

sd-coach turns Claude Code into a system-design interview coach for L3–L4 software roles. Choose a teacher-led coach, timed mock, Socratic tutor, rapid drill, or concept teach mode; sessions cover scoping, capacity, high-level design, APIs, schemas, low-level design, and failure modes. Mock sessions produce rubric-based scorecards, while Teach writes reusable notes to a personal concepts wiki.

GuitarTuner listens through the microphone, identifies the played note, and sounds a cue when the instrument is in tune. It is designed for guitar but works as a chromatic tuner for other instruments, with an adjustable reference frequency. The prebuilt macOS release is Intel-only; Windows users need to install the Python dependencies and run the program from the command line.

oss-contribution-harness gives Claude Code a forward-only workflow for deduplicating candidate work, studying repository norms, reproducing the issue, routing security findings, drafting the change, and planning follow-up. Every network-visible action requires fresh human confirmation, and the workflow also throttles contribution volume; it expects authenticated gh plus git and jq.

LaTeX Academic Publisher turns structured research into compilable papers using IEEE, MDPI, Elsevier, Springer, ACM, Nature-style, or plain templates. It scaffolds sections, manages BibTeX, places figures and tables, explains compilation errors, supports focused builds, and packages a clean submission archive.

Claude MD Architect helps Claude Code create a new CLAUDE.md, extract durable context from a long session, or audit an existing file that has become bloated or contradictory. Its workflows prioritize project-specific commands, rules, and gotchas that the model cannot infer, while moving longer material behind on-demand document pointers. Each mode uses a phased process and finishes with a fresh-session self-test.

Apple Skill gives agents practical AppleScript patterns for listing, reading, creating, completing, or deleting Apple Notes and Reminders. Because Voice Memos has no public scripting API, it documents the local SQLite schema and query patterns for listing or exporting recordings. It has no third-party dependencies, but the undocumented Voice Memos schema may change with macOS updates.

Give the skill a GitHub repository, project notes, or a plain-language description, and it reshapes the work for recruiters, hiring managers, and technical interviews. Outputs can include executive summaries, technical deep dives, Mermaid architecture diagrams, and STAR interview stories. It is aimed at engineers and architects who need to explain a project’s design and impact clearly.

rainforest-lab turns an open research question into competing directions, hypotheses, and candidate factors, then uses bounded gardener–skeptic deliberation and repeated gates to decide what survives. Useful failures remain as “golden leaves,” while lookahead or leakage results are blocked from reuse. The skill is shipped, but the Python engine and MCP product are still marked as in development.

openrouter-image-mcp gives Claude Code, Claude Desktop, Cursor, Codex, and other MCP clients one tool for generating or editing images through OpenRouter. It supports six aliases, aspect-ratio hints, automatic file extensions, multi-image saves, and clearer 401, 402, and 429 errors.

AlphaDesk feeds Claude a brokerage screenshot, CSV export, or MCP account view and produces a structured report covering holdings, market context, research, flows, risk, and P0/P1/P2 actions. It supports quick, standard, and deep modes, while leaving real orders to the user.

This Codex and Claude skill turns an article, topic, product screenshot, or reference cover into a title, composition, visual direction, and image prompt. It can also generate the cover, while sample analysis extracts reusable layout patterns without copying the original subject.

Paste Korean content to receive a ten-module GEO diagnosis covering evidence, quotations, citations, clarity, terminology, keyword density, and FAQ structure, followed by a domain-specific rewrite and comparison report. KoGEO instructs Claude not to invent statistics or quotes. Its method adapts research conducted mainly in English, so Korean results and changing generative-search rankings still need ongoing validation.