Recently updated open-source projects

img-deck guides Claude Code through content planning, editable white frames, a shared design system, slide prompts, image generation with kie.ai's GPT Image 2, and final PDF assembly. Two confirmation gates precede generation, while helper scripts support local editing, previewing, and packaging.

Use static navigation maps to locate official WeChat mini-program and mini-game guides, APIs, components, and services, then fetch a selected page into clean Markdown with headings, tables, and code blocks. The fetcher allows only official domains and retrieves page content on demand, so the repository stores an index rather than copied docs. Tencent retains copyright over the documentation.

Elucid changes Claude’s default response style so it gets to the point, explains technical terms in context, and teaches the reason behind each step and result. For news, new products, and other information that may have changed, it instructs Claude to check live sources and distinguish known facts from uncertainty. You can install it as a Claude Code skill or paste its instructions into a Claude.ai system prompt.

essnet-leave attaches to a Chrome session already signed in to an Orisoft ESSNet portal. It reads leave records and balances, compares them with the user's own Teams announcements, and can stage single- or multi-day applications; the form is not submitted unless --submit is passed. Chrome CDP, Playwright, and a live session are required.

Use Claude as an information architecture consultant for websites, apps, and content-heavy products. The skill can review navigation and labels, choose classification patterns, plan search and controlled vocabularies, and prepare card-sort or tree-test work. It is aimed at designers, content strategists, and agency teams that want practitioner-oriented guidance and client-ready audit outputs.

Give the skill a one-sentence startup idea and it researches customers, existing spending, failed attempts, risks, falsification tests, and a narrower defensible wedge. It then produces a designed HTML strategy document and scrollable pitch narrative with cited market research. A brutal, balanced, or encouraging tone can control the pressure level.

Run the setup script to check Go, Node, and Codex CLI, clone and build Moon Bridge, write a protected config with your DeepSeek key, back up the existing Codex config, and launch a local proxy on port 38440. A bundled Claude Code skill can guide the same setup, but the bridge still relies on the external DeepSeek service.

Provide a taxonomic scope, trait, seed papers, and optionally a curated dataset. The agent proposes a schema and project-specific validation hooks for approval, then searches, fetches, and processes literature in batches. A record reaches the CSV only after its quoted evidence and page are checked against the hashed source PDF, with provenance retained in an append-only ledger.

Analyze a JSON file, compare compact JSON with TOON, and convert only when the structure is likely to save tokens. Uniform object arrays tend to benefit, while deeply nested or heterogeneous data can grow. The CLI supports lossless JSON↔TOON round trips, uses tiktoken when available, and falls back to structural heuristics; TOON is mainly an input format.

Review public leadership communication before publishing, including social posts, job listings, all-hands notes, layoff announcements, and apologies. The skill returns a ship decision, quoted red flags, a ready-to-paste rewrite, the reasoning behind changes, and residual risk. It is not a blandness filter: it keeps forceful positions while removing contempt and avoidable self-inflicted damage.

Regulex+ turns JavaScript regular expressions into railroad-style diagrams in a browser or from the command line. It supports native Chinese and Unicode labels, bilingual UI, dark and light themes, PNG/SVG export, and text-based Mermaid output; the single-file page needs no backend and can be embedded. PNG and SVG rendering use headless Chromium, while Mermaid output does not.

AI-First CLI presents twelve principles for agent-facing command-line tools: structured output, clean stdout and stderr, actionable stable errors, non-interactive flags, raw API-shaped payloads, guardrails, token-frugal responses, and deterministic contracts. It includes a source-checked Vercel case study, self-audit checklist, and installable Claude skill. This is a design discipline, not a CLI framework.

Inspect an HTML deck in headless Chromium, capture computed geometry and styles, bake only hard-to-recreate decoration into slide plates, and place text, photos, and video as native editable PowerPoint objects. The scripts embed fonts and preserve browser coordinates, but each deck needs a few selector and font adjustments. PowerPoint honors the embedding; other viewers may need a PDF fallback.

Define products that combine physical hardware, firmware, apps, or cloud services without losing the boundaries between disciplines. Electrum guides a product idea through eight reviewed phases, producing system descriptions, component arrangements, gate checklists, diagrams, and presentation files. It works interactively through Claude Code, while the core templates and checklists can also be completed manually; a full guided run is designed to take roughly 60–90 minutes.

Configure Amazon SES to capture SMTP codes and remote-mail-server diagnostics through identity notifications or configuration sets, then relay clean notices with SNS and a Python Lambda. The skill generates an AWS CLI project, checks permissions, deploys only after confirmation, and verifies with the mailbox simulator. AWS credentials and cleanup of deployed resources remain your responsibility.

tudun-skill connects the 图盾·工坊 product-image service to agent platforms such as Hermes, OpenClaw, and Claude Desktop. A user supplies a product image and a natural-language brief, then the agent requests main images, selling-point posters, promotion visuals, or detail-page graphics and returns image URLs. It requires an API key and credits, so it targets ecommerce sellers and operators rather than general image hobbyists.

Agent OS Patterns documents practical shapes for products that need persistent state, memory controls, subagents, scheduled or proactive behavior, persona, and process isolation. The seven patterns cover filesystem-backed state, behavioral flags in memory, sibling sessions, linked work items and run directories, three-gate proactive checks, named failures with partial output, example-driven personas, and a localhost sidecar between UI and runtime. It is an architecture reference, not a prompt-only or library-debugging tool.

Use this Claude skill to question a fuzzy product idea one prompt at a time, focusing on who has the problem, what value they recognize, and what cost they would accept. Its value-consensus-model framework ends with a do, stop, or change diagnosis and a minimal validation experiment. The method is a synthesis, not a substitute for user research.