Recently updated open-source projects

Give reeljet real screen recordings, screenshots, a logo, and a brief to produce a directed short-form product ad. It creates the hook, storyboard, music choice, and estimated generation cost before an approval gate, then combines Higgsfield-generated motion with the real footage using captions and ffmpeg. The final campaign includes both 16:9 and reframed 9:16 videos and requires Claude Code, Higgsfield MCP access, Python, and ffmpeg.

ida-reverse-engineering drives IDA Pro through an MCP server to make binary databases and decompiler output easier to understand. The skill renames functions and variables, recovers structures, fixes types, adds comments, and follows an iterative cleanup loop with triage and library-resolution passes. Bundled IDAPython helpers cover strings, cross-references, crypto constants, fingerprints, and progress reports. It requires a connected IDA Pro MCP server and is intended for authorized reverse-engineering work.

claude-loop converts “make it better” into a project-specific scorecard and a repeatable cycle: score real signals, triage the weakest link, fix one slice, verify independently, and score again. Its init, score, run, and status modes keep the rubric and checkpoint state in .loop/. The loop stops after the target score is reached twice, with 90 as the default, instead of running indefinitely.

Litmus closes the gap between changing code and resolving a user’s problem. It drives the running application as the reporting user, follows the exact ticket steps, and inspects the live request and response, frontend state, and database. The tracker is updated only when the defect and its absence can be shown, so a green unit test, backend curl, or database row alone is not treated as proof.

research-reports combines a report-writing discipline with a research cache and a policy-source guide. It designs scope and terminology, stores full source text alongside URLs and local paths, grades evidence, declares boundaries, and reviews multiple sources before finalizing. The policy skill routes US sanctions, regulatory, and policy work to allowlisted domains and official APIs. It is for professional multi-source reports, not casual fact lookups.

Software Tailor Skills Library packages a custom-software consulting workflow into reusable Claude skills. Its modules cover niche research, discovery questions, project blueprints, value-based pricing, proposals, feature prioritization, onboarding, demonstrations, maintenance, bug triage, and follow-on work. The ZIP packages are meant for entrepreneurs using AI coding tools to serve B2B clients, including people without a traditional development background. Revenue figures in the examples illustrate the author’s method rather than guaranteed business outcomes.

gatekept verifies macOS application signatures with Apple’s codesign and spctl, flagging ad-hoc or unknown signers, injected dylibs, sideloaded apps, and Gatekeeper rejections. It also checks LaunchAgents, LaunchDaemons, Login Items, cron, shell startup files, temporary payloads, and hardening posture, then opens a self-contained local HTML report. The scan needs no account or dependency and does not modify the machine.

paypay-securities is a read-only CLI and agent skill for PayPay Securities accounts. It shows assets, cash, funds, US stocks, and trade history, measures stated fees and FX spread, and produces per-symbol summaries and review reports. Multiple accounts, Lark output, and separate session caches are supported.

wechat-typesetter is a reusable formatting specification for turning source text into inline-styled HTML that survives the WeChat public-account editor. It defines a two-color card system, eleven components, image-card wrappers, compatibility filters, and six self-checks. The source characters and paragraph structure must remain unchanged, with one output block per input paragraph. It targets WeChat on desktop and mobile; other publishing platforms are untested, and JavaScript is intentionally excluded.

This skill structures the academic lifecycle into nine stages and lets Claude work in mentor, reviewer, or mixed mode. It can help narrow a topic, plan literature and methods, challenge experimental design and reproducibility, improve a manuscript, simulate peer review, prepare a rebuttal, and shape submission strategy. Phase-specific references provide checklists, prompts, and reviewer red flags for researchers who want both direction and rigorous self-review.

Point the skill at a React or Next.js codebase to build a guided, theme-matched preview that visitors can click through before creating an account. It reconstructs only the required screens on in-memory mock data, adds a preview route and launch button, and audits the generated code to keep API clients, authentication, secrets, and real backend logic out of the demo.

moneyfrisk checks changed code for floating-point currency arithmetic, late rounding, and amounts written directly to a ledger. It points to the file and line, suggests integer cents or an explicit rounding policy, and keeps ambiguous tax decisions open for you to choose. Use it on checkout, billing, tax, balance, or settlement changes where a tiny numeric error becomes a real discrepancy.

cnipa-patent-writer assembles Chinese invention-patent abstracts, claims, descriptions, and figures from the user’s technical materials. When a reference DOCX is supplied, it borrows the layout and writing form without carrying over that patent’s technical content; without one, it uses a CNIPA-style fallback. The package can generate black-and-white figures and render pages for visual checking, but the user must provide an appropriate template and local document-rendering dependencies.

IRIS gives Claude Code a disciplined network-operations workflow for discovering device state, checking it against a source of truth, preparing a change, verifying the result, and documenting the ticket. Engineers retain the decision, while connected MCP tools, network documents, and discovery scripts supply context. Read-only automation can assist investigation; state-changing actions require explicit approval.

geo-content-forge audits a live page, classifies keyword opportunities from search signals, and writes paste-ready Title, Meta, H1, question-led H2s, FAQs, and FAQPage JSON-LD. It compares each element with the live page, labels facts by their source status, and blocks guesses, invented search volume, and ranking promises. A human-voice check also screens the output for common AI writing tells.

storyboard-prompter converts a one-line idea into one English prompt for an image model to render a multi-panel storyboard. It fixes the grid, aspect ratio, shared art direction, one beat per panel, and action logic so characters and style stay consistent across frames. The skill chooses common six- or nine-panel variants from natural language, but only produces prompt text; image generation happens elsewhere.

idea-discovery-framework guides a product idea through nine structured sections, scores commercial viability and solo-buildability against a consistent rubric, and returns a kill-or-proceed decision with explicit reasoning. It reads an idea brief, project instructions, and a builder capability profile, then writes each evaluation into a separate workspace.

motion-ui checks a project’s framework, package manager, Tailwind setup, and existing motion dependency before changing anything. It installs the appropriate motion package and builds 21st.dev-style components through Magic MCP or local recipes. Confirmation mode is the default; an opt-in auto mode runs end to end, and unsupported stacks receive an alternative instead of a forced install.