Start with the job, then find an open-source answer

Set up a three-document project record: a stable charter, a rolling work plan, and an append-only journal that records intent before risky operations. A lightweight heartbeat flags stale build markers, journal debt, and open questions, while session rules keep handoffs and code changes aligned. Existing files are offered for merging rather than overwritten.

Transform CSV, spreadsheets, JSON, XML, API payloads, or free-form documents into a JSON instance that matches a target schema, then report validation errors with their locations and keywords. The workflow asks batched questions when dates, enums, branching fields, or required values are ambiguous instead of silently guessing. The validator requires the jsonschema package.

Query Japan's post-1947 National Diet records by speaker, keyword, party, period, or session, then retrieve meeting lists, individual speeches, or full transcripts. The skill supports policy research, bill tracking, media fact-checking, and academic work without authentication. It still needs an HTTPS fetch tool, and Imperial Diet records before 1947 are outside its scope.

kuangjia-chaijie helps developers understand a cross-module framework topic before writing a talk or article. It combines top-down mapping with bottom-up symbol tracing, scope checkpoints, phase exit criteria, and clean-context cross-checks, then produces structured Markdown notes rather than the article itself. It is not intended for single-function debugging.

Bump Java Version detects the effective bytecode target across a Maven or Gradle project, then performs one hop such as 8→11 or 17→21. It records the old-JDK test baseline, applies OpenRewrite and build adjustments, and checks compilation, retained tests, and the new target under the new JDK. Each hop needs both JDKs, the project build tool, and Maven Central access.

Key Message Review examines prescription-drug marketing messages for empty claims, weak strategic grounding, and common compliance risks, then rewrites them around a concrete patient, a memorable point, and a clinician action. It offers quick, standard, and deep modes plus an optional before-and-after card. It is not a substitute for medical or regulatory sign-off and should not invent efficacy data.

Moto News is a Claude Code plugin for drafting content about Harley-Davidson and Indian Motorcycle. It covers new-model news, motorcycle reviews, event coverage, riding culture, and social posts for channels such as Facebook, Instagram, and X. The repository describes content-generation prompts, not live news retrieval or automatic publishing.

Write AI News turns a supplied URL or news title into Thai AI-news copy: a headline of up to 60 characters, a one-sentence summary, three to four body paragraphs, tags, and source information. It is a formatting and drafting skill for Claude Code; the project description does not promise automatic live-news retrieval or verification.

Chinese EPUB Typo Fixer handles OCR mistakes in Chinese ebooks made from scanned images. It applies clear visual-character or simplified/traditional corrections by default, lists archaic-usage modernization separately, and outputs a corrected EPUB plus a change report. The scripts only unpack, rewrite, and repack; they avoid network access and skip ambiguous or oversized changes.

Vastu-Jyotish Mentor uses birth and business details to calculate charts, planetary periods, auspicious times, Vastu considerations, and numerology, then returns a short verdict with plain-language reasons or suggested remedies. It uses pyephem and keeps personal data in a local private profile. The project frames this as a traditional cultural lens, not a scientific guarantee or a basis for medical, investment, or marriage decisions.

Humanamente rewrites Spanish AI text by identifying inflated phrasing, negative parallelism, mechanical lists, English punctuation, generic conclusions, and other local writing tells. It preserves meaning while iterating through a six-dimension style check, without regex, scripts, or find-and-replace rules. A simple mode is available, with optional voice profiles for close, corporate, technical, or commercial tones.

Holo-Zlib is a portable Agent Skill for searching and downloading ebooks through the Z-Library eAPI, from Claude Code, Codex, Cursor, Gemini CLI, or a shell. Runtime use needs only Python 3.10+ and the standard library. Credentials and downloads stay outside the skill directory, but users must prepare credentials/zlib.json and can choose an output directory.

TikZ Scientific Figures turns a plain-language description into editable TikZ or PGFPlots source, compiles it into a vector PDF, and returns both files. It covers plots, circuits, biological and physical diagrams, and other technical figures, with optional SVG editing through bundled tools. A LaTeX toolchain such as pdflatex is required, with extra dependencies for previews and SVG export.

Circuit reads Gmail in a read-only flow, keeps threads with genuine two-way replies, filters newsletters and automated senders, and surfaces relationship type, status, deadlines, and action items. The open-source version can use Claude's Gmail connector or a local MCP engine, with no Circuit server handling the mail. Background sync and a web dashboard belong to the hosted tier.

Code Inspector is a Go CLI and library for finding worthwhile refactoring targets across many programming languages. It combines complexity, git churn, duplication, maintainability, and dependency-graph metrics, then reports file- and function-level details plus ranked hotspots. The repository also includes a Claude Code skill for first-pass code-health analysis. Non-Go language support requires cgo and a C compiler.

youtube-learn turns a YouTube link into structured notes for lectures, podcasts, interviews, and tutorials. It fetches captions with yt-dlp, falls back to Whisper when needed, reads the full transcript, and writes a TL;DR, key takeaways, topic-by-topic explanation, notable quotes, and audience guidance. Tutorials are organized like reference material, and the result is saved as Markdown.

autoresearch-skill gives Claude a disciplined loop for optimizing a measurable target such as test runtime, bundle size, latency, memory, or training loss. Each experiment is benchmarked, useful changes are committed on one branch, and regressions or failures are reverted while the run history remains available. Session files support resuming later, but the approach is intended for iterative work rather than trivial one-off edits.

ziyu packages crisis-communication playbooks for brands, creators, and small teams using Claude Code. It covers eleven crisis types with severity assessment, information checklists, response strategies, action lists, recovery timelines, and templates for apologies, clarifications, denials, and legal warnings. Separate monitoring guidance supports current-issue scans and scheduled checks that organize signals into a structured public-opinion report.