Recently updated open-source projects

Give the mapper a brand or product brief and it fills only the missing context about audience, mood, and platform. It then resolves a design route across color, visual effects, typography, and icons, carrying decisions from one specialist skill into the next. You can ask for the route alone or let it apply the sequence automatically; the full workflow depends on its companion skills being installed or fetched.

NotFair is a Claude Code marketing plugin for investigating wasted ad spend, traffic losses, creative fatigue, and SEO issues through Google Ads, Meta Ads, Search Console, and related connectors. It can perform only the account changes exposed by those connectors and also includes a local CMO portal with specialist agents. Connected ad features require OAuth, and the Meta write surface is intentionally limited.

humanize shapes prose as it is generated or cleans an existing draft while preserving quoted text, citations, numbers, and proper nouns. It adjusts style for the document’s register, can learn a reusable voice profile from writing samples, and includes a standard-library Python linter for CI. The linter detects mechanical patterns and density; it does not prove whether a person or model wrote the text.

Chatd is a desktop application for asking questions about local documents with a Mistral-7B model. Document data stays on the computer, and the app bundles the local model runner so users do not need to assemble a separate stack. If Ollama is already running, Chatd reuses it; otherwise it starts and manages its own Ollama service. Users can simply download and run the executable, then optionally enable GPU support or choose another model.

SvelteStorm brings editing, a project tree, terminal access, browser preview, and component-state inspection into one desktop workspace for Svelte development. Its performance dashboard, split view, search, linting, and snapshot-based time-travel debugger help developers follow how interactions change application state. It targets macOS and Windows, while running the IDE from source requires the documented Node.js version and a more involved set of build prerequisites on Windows.

React NodeGUI is a React renderer for NodeGUI that builds Qt5-based desktop applications for Windows, macOS, and Linux. Developers can compose native widgets with React, style them with cascading CSS and Flexbox, and use Node.js APIs without bundling a Chromium runtime. The repository explicitly states that the project is no longer maintained, so adopting it now means accepting an aging stack or maintaining a fork yourself.

Readme.skill analyzes local usage records from Claude Code, Codex CLI, Kiro, Cursor, and other supported tools to create an AI-collaboration profile and a vertical SVG poster, with optional monthly or custom date-range reports. Processing is read-only and local by default, and project names, private repositories, and secret-like strings are anonymized; only the optional GitHub contribution lookup uses the network. Because conversation content may be analyzed to infer working patterns, users should still review the generated files before sharing them.

DingTalk Wukong Skills is a multi-skill collection extracted from Wukong v0.9.2 for Claude Code, Gemini CLI, Codex, and similar agents. It covers DingTalk OA, tasks, calendars, and documents; Word, Excel, PowerPoint, and PDF handling; and travel or local-information queries for 12306, Ctrip, and Dianping. A skill-creation utility is included. Because the repository nests skills across several directories, installation must use full-depth discovery. The package reflects a specific Wukong release rather than a live DingTalk client.

zot packages a coding-agent loop into a single Go binary with interactive TUI, print, and JSON modes plus read, write, edit, and shell tools. It supports many direct APIs, selected subscription logins, cloud platforms, and local Ollama-style models, with session recovery and cost tracking. Optional subprocess extensions can be written in any language, and a built-in Telegram bridge enables remote conversations.

AgentRL separates agent reinforcement-learning training from environment deployment. Rollout, actor, and reference workers run on Ray to generate trajectories, update the policy, and maintain a frozen KL baseline, while a controller and task workers manage multi-turn environments over HTTP or gRPC. It is suited to reproducing the paper’s experiments or adding tasks, but training requires Ray, suitable CPU/GPU resources, and model configuration.

Zeroshot orchestrates an agent you already use through an executor-verifier loop. One agent writes the change, while an independent verifier checks the result without seeing how it was produced and can request another pass. It supports Claude Code, OpenAI Codex, Gemini CLI, and OpenCode from a command-line workflow.

cuecards assigns or auto-detects a profile for each project directory, then launches Claude Code or Codex with only the relevant skills, MCP servers, plugins, persona, and quality gates. The same profile can also be materialized into native configuration files for Cursor, Cline, and other supported tools. It reduces irrelevant always-on context and global configuration sprawl, requires Node 20 or newer, and acts as a configuration and launch layer rather than an agent replacement.

brand-launchpad turns a project or person into twelve actionable brand files covering the brand journey, positioning, hero line, handles, formats, recurring series, content mix, idea sources, a 30-day plan, and operating rules. It selects a brand variant first, then applies practical writing and strategy frameworks.

Agent Sessions is a local-first macOS app for browsing and searching histories from Codex, Claude, OpenCode, Cursor Agent, Hermes, OpenClaw, Copilot CLI, and other supported tools. It shows transcripts and images, saves sessions, and can open resume commands in Terminal, iTerm2, or Warp. Session data stays on the Mac with no telemetry or history uploads; macOS 14 or later is required.

Inspect Evals gathers benchmarks for coding, cybersecurity, safeguards, reasoning, knowledge, and other model capabilities. Researchers can run individual tasks with Inspect AI, improve included evaluations, or register externally maintained implementations. The project recommends Python 3.11 or 3.12 for the most reliable development and execution experience.

Loki Mode turns a PRD, issue, API specification, or short brief into a repository with source, tests, configuration, and audit records. Its agents repeat reason-act-reflect-verify cycles, while deterministic gates block empty diffs, failing tests, and inconclusive checks from being reported as complete. Each run can produce evidence that the completion claim can be rechecked.

osint-field-graph defines a field and disqualifier, seeds an initial set from GitHub topics, OpenAlex, or a supplied edge file, then expands the graph with forward BFS and ranks nodes across sources. Persona cross-checking and a domain filter narrow the result. The core is pure shell, and GitHub mode needs authenticated gh. Its scope is limited to public accounts, public relationship graphs, and signed work; it is not a deanonymization tool.

wp-elementor-mcp-skill turns a documented product JSON spec into Elementor’s _elementor_data widget tree and writes it through WordPress REST or MCP. It includes endpoint checks, offline structural validation, and a dry-run path, with real builds defaulting to draft pages. After an API write, update the page in Elementor or regenerate CSS and data so its cache picks up the layout.