Recently updated open-source projects

AutoHarness is a lightweight governance layer for existing AI agents. It can wrap an LLM client, check tool calls for risk and permissions, sanitize outputs, record decisions, and manage context and token budgets. YAML rules select different governance levels, while multi-agent profiles, per-call cost attribution, and session persistence cover longer-running workflows.

MeshStation is a desktop Meshtastic observatory that passively receives and decodes live radio traffic through an RTL-SDR-compatible device. It maps nodes, displays messages and raw packets, builds a node database, and helps evaluate coverage, density, reachability, and network quality without requiring a Meshtastic node. The current application is focused on reception and analysis; transmission and active network interaction are planned rather than available today.

Letta Evals organizes agent evaluation from dataset and target through extractors, graders, rewards, and stored results. It supports JSONL or CSV data, multi-turn samples, multiple model handles, cached re-grading, deterministic or model-judge graders, and custom agent factories. Runs target Letta Code through a self-hosted or cloud server, so Python 3.11+, a running Letta service, and provider credentials are required.

Install this skill pack when an AI coding agent needs to design or refine motion rather than merely add an animation. Nine focused modules cover timing and easing, color, composition, art direction, After Effects, beat-synced editing, Remotion video, logo animation, and looping backgrounds. The pack works across Claude Code, Cursor, Codex, and other compatible agents.

dig helps coding agents avoid building the wrong thing from a vague request. It surfaces assumptions about the goal and likely failure points, asks only questions that would change the approach, loops until important branches are resolved, and produces a concise clarity memo for confirmation. It follows the Agent Skills format; Claude Code offers the fullest interaction, while compatible clients can use a reduced workflow.

Provide product names, ingredient text, or a label photo to review a vitamin or supplement stack against intake targets, upper limits, duplicate ingredients, and known interactions. The Claude skill links claims to sources such as NIH ODS, PubMed, Cochrane, and Korean regulatory data, with optional product lookup APIs. It is designed for adults and does not diagnose conditions or advise changing prescription doses.

Social Media Content Planner takes a client from onboarding to a dated monthly calendar. It collects brand and audience constraints, organizes research and content in a Google Spreadsheet, and researches current topics through web search, with the YouTube Data API available as an optional enhancement. Explicit approval loops run in Slack or chat before the calendar is handed to production.

HoYo.Gacha is an unofficial Windows tool for managing multiple accounts and gacha records from Genshin Impact, Honkai: Star Rail, and Zenless Zone Zero. It reads Chromium disk cache data and contacts the relevant API without requiring a local proxy, then stores the results in a local database. UIGF and SRGF are supported for import and export, while CSV is export-only. Because the database contains all local accounts and history, it must be preserved when moving the application or migrating the operating system.

paper-explainer reads the full paper, including appendices, tables, and figures, then builds an anchored knowledge graph for a structured explanation. It covers positioning, limitations, methods, experiments, critical analysis, and future work, with quick, full, and reproduction depth levels. Follow-up questions return to the cited section before answering, making it a comprehension tool rather than peer review.

Hydra is a Claude Code skill for reviews that benefit from several independent perspectives, such as architecture decisions, security-sensitive changes, and difficult refactors. Standard mode uses four advisors plus a chairman; deep mode expands to six advisors, three reviewers, and optional Codex participation. The final report traces recommended actions back to finding IDs and asks for cost confirmation before running.

de-ai-writing-skill rewrites supplied prose by replacing connective em dashes, not-X-but-Y constructions, LinkedIn vocabulary, stacked aphorisms, and manifesto-style endings. It keeps the original specifics, dates, numbers, and personal register, and does not add or remove claims. Use it for emails, submissions, blogs, landing pages, and other copy judged by people, not for code comments or internal notes.

Manage Linear work without leaving Claude Code: create projects and sub-issues, update statuses, apply labels, and sync batches of code changes. The skill supports simple MCP actions alongside TypeScript SDK scripts and direct GraphQL queries, so teams can move from routine commands to more customized automation. A Linear API key is required.

excel-house-charts creates native Excel line and bar chart objects with a fixed style for fonts, country colors, axes, legends, gridlines, and labels. Its helper distributes line-chart end labels and adds leader lines, while bar charts use a defined gap width and rotated value labels. openpyxl is required, and Gotham plus actual Excel are recommended because LibreOffice may not render manual label placement correctly.

findable searches current SEO, GEO, and llms.txt guidance, reads the project’s stack, and can apply low-risk changes such as robots.txt, sitemaps, meta tags, JSON-LD, canonical links, and missing alt text. Sensitive edits are recorded in TODO SEO.md with the source URL, exact file, and rationale for you to decide. It works across common static-site and web frameworks, but it relies on current web research.

Describe the telemetry flow you need or paste a broken Collector file, and otelcol-doctor will produce a commented configuration or explain the fault and repair it. The skill consistently checks processor ordering, core-versus-contrib components, pull-versus-push exporters, disconnected pipelines, and other common failure modes, then runs otelcol validate when available. It focuses on Collector configuration rather than application SDK instrumentation or dashboards.

This template combines Tauri 2, the Next.js App Router, Tailwind CSS, and TypeScript with preconfigured Biome, Rust formatting checks, and GitHub Actions. Developers can run the frontend inside a Tauri window and package a statically exported build for release. Because a packaged Tauri app does not include a continuously running Node.js server, the frontend must use static site generation rather than Next.js server-side rendering. The author also notes that Next.js may be more framework than many Tauri apps need.

OSOP Agent Rules installs platform-specific instructions for 18 coding-agent environments, including Cursor, Codex, Windsurf, Aider, Cline, and Claude Code. After a multi-step task, the agent writes an `.osop` workflow and an `.osoplog` execution record containing nodes, dependencies, status, duration, and tool usage. The files can be viewed in the OSOP visual editor or used to generate reports.

The builder interviews you about role, topics, channels, tone, and hard style rules, then analyzes 5–20 real samples or a Telegram export for recurring phrases, rhythm, and signature markers. It generates an invokable Claude Code skill plus a reusable voice profile that can be updated with new writing later. The analyzer uses only Python’s standard library, but sparse or overly polished samples produce a less faithful result.