Recently updated open-source projects

This Claude Code skill reviews AWS spend through the CLI and Cost Explorer, breaking costs down by service, region, and resource while explaining usage and recent activity. It separates reversible savings from larger changes that need confirmation, stays read-only by default, and can create a self-contained HTML dashboard.

Apply a seven-stage methodology that takes tabular data from cleaning and exploration through feature engineering, modeling, diagnosis, and an interpretable report. The workflow emphasizes leakage checks, baseline comparisons, cross-validation, and honest failure analysis rather than chasing a single score. It can run as a Claude Skill or as a portable prompt for other language models, with output adapted to the user's language.

Use this Claude Code skill to compose melodies, chords, bass, drums, full arrangements, or simple sound effects with local system tools. MIDI is rendered through FluidSynth or Timidity and a General MIDI SoundFont, while direct synthesis handles tones and sweeps. The process is deterministic rather than text-to-music AI, and MP3 output additionally needs ffmpeg.

Fill flat PDFs, comb fields, AcroForms, and some XFA forms locally, then drag values over the rendered page or nudge them by one pixel before export. The tool keeps documents on your machine, leaves signature lines blank, and can use a private profile for autofill while skipping SIN, bank, and card numbers. It does not require Adobe or a cloud service.

Crack the Paper turns papers, technical reports, long-form articles, PDFs, and arXiv links into structured reading notes. It follows figures or claim clusters, questions methods and alternative explanations, tracks later work and reception, and connects the material to the reader's research or courses.

claude-code-colorize-project assigns each VS Code workspace a distinct frame color, making parallel Claude Code sessions easier to recognize in task switching and window previews. Choose from 20 dark presets or provide a custom hex value; the skill updates both folder settings and detected multi-root workspace files while preserving unrelated configuration. Editor and chat content remain neutral so readability and syntax highlighting are not changed.

Create structured handoff documents containing state, decisions, next steps, critical files, and gotchas, then validate or resume them in a later session. This fork resolves references from detected repository markers, accepts any ATX heading level, adds project-root and minimum-score flags, and includes regression tests. Validation also checks placeholders, missing files, and possible secrets.

Use sprint-orchestrator to keep one Claude Code chat focused on planning and review while separate sprint chats implement features in isolated Git worktrees. Each sprint can open a PR and preserve decisions and recurring bug lessons in repository state, reducing context growth across long projects. It is aimed at ongoing, multi-feature development rather than one-off scripts or tiny fixes.

This skill turns Claude into a demanding editor for essays and articles. It scaffolds core paragraphs instead of writing the author’s position, surfaces one central drift or author-specific trap per turn, and records the thesis, strengths, decisions, and progress in a persistent tracking file for the next session.

This Claude Code skill curates about 40 Paul Graham essays across high agency, startup ideas, founders, growth, fundraising, product, and mental models. Trigger rules surface the relevant frame during a founder conversation, with short fair-use excerpts and canonical links, but no current news or legal advice.

Obsidian Study Guide Builder turns lecture PDFs, study guides, and notebooks into an interlinked vault of concept notes, formula walkthroughs, lecture teaching notes, visuals, animations, and spaced-repetition cards. It asks about scope and build cadence, shows a sample, then generates and verifies the rest.

skild is a package manager for Agent Skills that installs from Git URLs, local directories, or its registry. It can list, update, validate, remove, and synchronize skills across platforms including Claude, Codex, Cursor, Windsurf, and others, at either global or project scope. Users can also install bundled skillsets, initialize new skills, push them to Git repositories, and publish them through the registry.

Upload a paper, thesis, dataset, or analysis script to receive stage-aware feedback on structure, argumentation, and internal consistency. The skill can plan required sections and assets, then cross-check figures in the manuscript against Python, R, SPSS, Stata, or MATLAB outputs. It is designed to guide students through research decisions rather than ghostwrite the final work.

Give the skill a claim and public data, and it checks nine common pitfalls including Simpson’s paradox, survivorship bias, relative-risk inflation, base-rate errors, selection bias, and uncorrected multiple comparisons. It produces Markdown, journal-style HTML, and a dashboard PNG while retaining supported positive scenarios. It is a warning tool, not professional medical, statistical, or legal advice.

Bring a concrete leadership problem such as team structure, hiring, underperformance, executive alignment, or a VP transition, and receive a diagnosis, a relevant framework, a recommendation, explicit trade-offs, and anti-patterns to avoid. The skill covers organization design, managing up, cross-functional relationships, strategy, performance, culture, scaling, and self-management for senior design and product leaders.

PyTrace Autopsy gives Claude Code runtime visibility when a failing Python test involves state mutation, swallowed exceptions, or incorrect values moving through a long call chain. It injects a temporary `sitecustomize.py` through `PYTHONPATH`, uses `sys.settrace` to record call, return, and exception events, and writes a reconstructable JSONL call tree. The approach requires Python 3.8 or newer and leaves no files or modifications in the target project.

Define what better means in program.md, then let the agent make one change per experiment and score it against a frozen evaluator. Improved artifacts are committed; regressions and crashes are reverted, with every attempt logged in results.tsv. The workflow covers code, prompts, content, models, and configs, but the evaluation boundary must stay immutable.

Build-DNA begins with an adaptive interview about a user’s skills, interests, network, location, resources, and motivations, then proposes business ideas matched to those advantages across software, physical products, services, commerce, and creator businesses. For a chosen idea, it develops market and founder-fit analysis, business-model options, first-customer channels, an MVP roadmap, and a concrete 72-hour action plan. It is aimed at builders and founders who want a direction grounded in their own circumstances.