Recently updated open-source projects

Eclaire unifies tasks, notes, documents, photos, and bookmarks in a local-first workspace where users can search, classify, OCR, and act on their content through an AI assistant. It can run in one container with SQLite or PostgreSQL and supports several local model backends. Rich document and image processing requires additional system tools beyond the core deployment.

ai-security-guard combines zero-dependency TypeScript runtime guards with a Claude skill for prompt injection, replayed-history attacks, privilege escalation, forged action markers, cross-tenant PII leaks, media abuse, and unsafe rendering. It can log blocked attacks and propose new patterns, but each candidate must pass a regression check before a human promotes it, so nothing becomes live automatically. The reference code uses Express-style functions that can be ported to other stacks.

Butler treats the main agent as the execution owner: it understands the goal, makes key judgments, integrates results, and performs final acceptance. It may delegate bounded, verifiable subtasks to a lower-cost model only when the total effort is worthwhile, and it can choose to do the work directly instead. Delegated output must be checked by the main agent, with explicit terminal states for completion, decisions, or a new work window.

This installable Agent Skill gives coding assistants focused guidance for Playwright testing in TypeScript. It maps the current task to references for end-to-end, component, API, visual regression, accessibility, security, mobile, multi-tab, PWA, Electron, browser-extension, and internationalization testing. Other material covers CI, sharding, reporting, performance, and flaky-test debugging. Installed through `npx skills add`, it helps an agent choose suitable patterns; it does not replace Playwright or run tests by itself.

Kode is an open-source terminal assistant with multi-model switching, expert consultation, specialized subagents, code editing, command execution, MCP extensions, AGENTS.md discovery, and an ACP server mode. It supports interactive, non-interactive, and plan workflows, plus plugins and model configuration files. Its default mode skips most approval prompts, so safe mode is better for sensitive work. The repository also says a major maintenance and refactoring phase runs from July through December 2026.

Start with one launch, demo, build log, lesson, or release, then define the weekly narrative, audience order, calls to action, and daily assets around it. The framework suits builders, founder-led brands, and small content teams, and works either as a Claude Code skill or as a plain Markdown template.

Turn git push into a lightweight deployment loop for an agent, bot, scheduler, or scraper running on one VPS. A Python standard-library webhook listener verifies GitHub payloads with HMAC, fast-forwards the repository, and stays running through systemd. It is not a build system or multi-host deployer, but a lightweight sync loop for one host.

Follow a deployment runbook from DigitalOcean droplet provisioning through Python 3.11 setup on Ubuntu 24.04, non-interactive Hermes installation, profile and workspace linking, credential migration, and OAuth callback caveats. It finishes by running the gateway as a lingering systemd user service for an agent that needs to stay online.

ToolGen represents tools as dedicated vocabulary tokens so a model can select tools, call them, and continue generating language within one process. The repository includes models, data for its three-stage training procedure, and evaluation code for research reproduction. Its local inference example also requires a ToolBench key and a StableToolBench deployment.

oss-launch scans a repository, asks only for missing project details, and generates a tailored open-source file set such as a license, README, contribution and security policies, changelog, CI, and GitHub templates. Its scripts can build a clean package and re-check gaps, while optional launch guidance covers metadata, releases, demo media, and community posts.

Kasetto is a declarative CLI for keeping AI-agent environments consistent across projects, machines, and teams. A YAML file and lockfile define skills, commands, MCP configurations, and instruction files, which can then be synchronized into Claude Code, Codex, Cursor, and many other agents. It also supports composable configs, private repositories, locked CI runs, and secret injection from existing credential managers.

Give a coding agent a video brief and use these skills to move from script and storyboard through scene construction, narration, captions, editing, and rendering. The pack includes workflows for short explainers, data-driven vertical recaps, progressive diagram and chart reveals, whiteboard drawing, and isometric scenes. It works with Claude Code, Cursor, Codex, and many other agents.

zuiti-huabing reads workplace promises such as “take this on now and promotion comes later,” separates encouragement from cost shifting, and produces three reply styles: polite, pointed, and firm. Each response aims to clarify the result, timing, and written record without insults, threats, retaliation, or deciding whether you should stay. It is a scenario-specific package within the broader 嘴替.skill ecosystem.

zuiti-jiafang handles client messages about discounts, extra requests, delayed payment, or restoring an earlier version without moving the deadline. It explains the implied cost transfer, then offers polite, pointed, and firm replies organized around scope, timing, definitions, and value. Its firmest advice stays within documentation, contract, and lawful remedies; it does not encourage threats, public shaming, or retaliation.

zuiti-cuihun interprets marriage questions by distinguishing comparison and pressure from genuine parental anxiety. It offers polite, pointed, and firm replies that acknowledge care while keeping your timeline and choices yours. The package is designed for family messages and conversations; it preserves workable boundaries without assuming every relative is acting in bad faith.

This Claude Code skill audits repositories, database schemas, APIs, and features against Chile’s Law 21.719. It inventories personal data, checks 13 control areas, ranks code-level gaps, and produces a versionable report with file-line evidence, legal references, and remediation priorities. A focused mode covers one feature or pull request; the result is an engineering audit, not legal advice.

Pixel Surgeon offers Build, Operate, and Deslop modes for designing, reviewing, or tightening interfaces. Its searchable base contains 504 practices and 40 style recipes, while a 12-point exam covers hierarchy, typography, color, spacing, controls, states, and accessibility. It can inspect screenshots, URLs, React components, or Figma frames and return fixes with exact values. It requires Python but no extra library dependencies.

product-design-partner first identifies one of nine lifecycle phases, then routes the work through strategy, discovery, design, validation, shipping, measurement, growth, AI UX, or portfolio modes. It produces concrete artifacts such as assumption maps, interview guides, usability plans with sample-size math, positioning, and rollout metrics. Sourced numbers are cited or labeled illustrative, making it more actionable than generic product advice.