Recently updated open-source projects

seo-chips-internal-linker turns category and optional brand URLs into a two-sheet Excel map of related-page chip links. It mixes child, sibling, parent, complementary, and occasional brand targets, then uses DataForSEO monthly volume to validate anchor terms and adds a note when an anchor differs from the page name. Without that optional service, the workbook still builds but unknown volume cells remain blank.

Watson turns a brief field observation into three distinct takes: Service Design, Business Model, and Trend. Each lens adds one Socratic question, with an optional deepen mode for a fuller analysis. Saving is opt-in and writes a structured Markdown note only after the user asks, so the default experience remains chat-only.

Layerdex uses hfviewer to render a public Hugging Face model at a chosen granularity and save a cropped architecture graph plus normalized JSON metadata. The JSON can include nodes, operation types, vocabulary, and attributes, and is checked against a published schema. It needs Playwright and Chromium, does not download weights, and does not support gated or private models.

netis-nx62-openwrt-skill walks through the U-Boot TFTP recovery path for the Netis NX62 and Netcore N60 Pro, including image checksums, flash backups, live partition discovery, and packet-capture gates. It encodes board-specific recovery pitfalls and orders writes so the irreversible step is delayed. Flashing can brick the router, so the guide requires live verification and explicit user risk acceptance.

paxel-skill evaluates local Claude Code and Codex CLI sessions against Paxel’s five axes: execution leverage, steering, engineering quality, product thinking, and planning. It reads project logs and Git history, removes tool bodies and secrets, extracts events and decisions, then asks Claude Haiku for narratives and scores without uploading data or using a backend. Results include a performance band, but the project warns that the model differs from YC’s scorer, reruns vary, and the overall score is only an approximation.

Body Root Cause Skill uses questions, a persistent health record, and Bayesian updates to organize possible causes behind skin, fatigue, sleep, and related concerns, then separates root, trigger, and maintaining factors. It tracks exams, medicines, supplements, and outcomes across conversations, while staying focused on oral supplements. It does not diagnose; worsening or persistent symptoms still call for medical care.

LionClaw wraps existing agent CLIs such as Codex and OpenCode in a project-local control plane. It keeps runtime profiles, durable sessions, skills, channels, scheduled jobs, private state, and audit records under a `.lionclaw/` directory while preserving the underlying agent’s own interface. Use it when you want swappable runtimes with explicit control over workspace access and credentials.

Swifty is an offline-first password manager for macOS, Windows, and Linux. It keeps login credentials, credit-card details, secure notes, generated passwords, and time-based one-time codes in an encrypted file on the local machine, unlocking them with a master password. Backups can move the vault between computers, while Google Drive synchronization is optional for users who want it. Packaged releases are available, or the desktop app can be built from source.

Components lets an agent describe an effect such as a scroll-driven laptop, tilt card, or globe, fetch the matching React and Tailwind component from a live library registry, and adapt it to the project’s design tokens. It keeps pointers to maintained upstream code so updates and licenses travel with the source, and includes plain fallbacks plus guidance for responsive and reduced-motion behavior. The workflow needs a project that can install the selected dependencies and access the source registry.

Harbour connects iOS, iPadOS, and macOS devices to multiple Portainer servers and endpoints. Users can inspect container health, environment values, mounts, networks, and logs, change container state, remove containers, and deploy or edit complete stacks. Widgets, Shortcuts, Handoff, Spotlight, and background notifications extend the Apple-platform workflow. A working Portainer installation is required, and creating or editing individual containers is not currently supported.

Cami Design reads project context from a `.cami.md` file and routes each issue to a focused layout, interaction, copy, or engineering mode. The visual modes review hierarchy, motion, states, labels, and tone, while the engineering pass checks composition, design-system fidelity, accessibility, performance, state handling, and types before handoff.

Oxide Lab provides a desktop chat interface that runs compatible language models entirely on the user's machine. It loads GGUF or SafeTensors models across several Llama, Qwen, Mistral, DeepSeek, and related architectures, with CPU, NVIDIA CUDA, or Apple Metal acceleration. Users must obtain a compatible model before chatting, and larger models require more memory than the stated 4 GB minimum.

Use this Claude skill to design recommendation, feed, search-reranking, or task-prioritization systems as six stages: Source, Hydrator, Filter, Scorer, Selector, and SideEffect. It walks through the trade-offs between candidate sources, filtering, scoring, selection, and asynchronous effects, then emits a runnable scaffold in TypeScript, Go, or Python. Examples cover feeds, a Go module, and a FastAPI task ranker.

CardZero creates a USDC wallet on Base for an AI agent, while a human owner claims it and configures per-transaction limits, daily caps, address whitelists, and emergency freezing. Agents can check balances, pay for services, or use job escrow through a REST API, MCP server, and platform-specific integrations. The product is still in beta, its smart-contract audit is in progress, and the repository recommends keeping balances below 100 USDC.

product-prover reviews PRDs, feature specs, architecture documents, and design proposals before coding. It extracts states, transitions, actors, boundaries, invariants, safety, liveness, and rollback conditions, then ties each finding to a source location, trigger, consequence, and concrete fix. Unsupported conclusions stay marked as missing specification rather than invented. Full-spec and focused cross-link passes are available, with findings persisted by status; this repository is a read-only mirror of a larger pack.

Ask an agent to add temporary alternatives to a rendered surface, switch between them with a local picker, and name the visible winner in chat. Unship removes the losing options and checks that temporary artifacts are gone before release. It has no telemetry or remote session store, and is intended for local comparison rather than production experiments, analytics tests, auth, or payment flows.

box-rclone-binder uses Box server authentication so each host mints its own short-lived access token instead of sharing a rotating OAuth refresh token. The box-binder CLI deploys one Box drive to multiple servers through rclone, runs read-only health checks, performs bounded refresh or self-heal, and can schedule checks with cron or systemd plus Discord alerts. Fleet configuration stores pointers to secrets rather than secret values.

The skill captures project-related thoughts during conversation, preserves the user's original wording, and organizes them into a project-bound mdbase. Later sessions or different agents can read that repository to inherit requirements, preferences, and constraints instead of reconstructing them from scratch. Commands are provided to review, sort, ignore, and export records, and the host agent needs file read/write and Bash access.