Claude Skills
ojura/claude-skills
Skills prontas para produção: gestão de conversas, automação remota e análise de custos
Ver no GitHubColeção de skills desenvolvidas para Claude Code que resolvem problemas práticos do dia a dia: gerenciamento de conversas, monitoramento de custos de cache, controle remoto de máquinas via HTTPS, transformação de documentos em PDF e execução de comandos administrativos. Inclui ferramentas para recuperação de sessões deletadas, filtragem de conversas e abstração de raciocínio visível.
Instalação
npx convez-ai ojura/claude-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Recuperar conversas deletadas do Claude Code
- Controlar máquinas remotas via tunnel HTTPS em redes restritas
- Organizar e arquivar automaticamente histórico de projetos
- Monitorar custos de prompt cache e otimizar uso
- Visualizar raciocínio e thinking do modelo no fluxo principal
- Executar comandos administrativos locais com segurança
Veredito

Conjunto sólido e funcional com skills maduros para casos reais: especialmente valiosas são a recuperação de sessões ext4 (prática salva dados) e o remote-shell-over-chisel (engenharia elegante de redes). Exigem ambiente técnico robusto e familiaridade com linha de comando, mas pagam dividendos em automação e segurança.
README
A tradução para o português deste README ainda está na fila.
claude-skills
Skills I built for Claude Code and use myself: tidy and recover your conversations, see what your prompt cache is costing you, pull your own data out of Claude and ChatGPT, turn Markdown or HTML into print-quality PDFs, and drive your real browser or let Claude run admin commands on your machine, safely. Each one lives in its own folder with its own README.
Reviewing and fixing branches
review-decision-maker
Takes a pile of reviewer claims through current-code verification, explicit Fix, Discuss, Defer, or Reject decisions, implementation, tests, and a final outcome ledger. It includes a reusable local HTML board with filtering, notes, persistence, import, JSON and CSV export, and standalone HTML snapshots with current decisions embedded. Read more.
Working with your conversations
archive-conversation-forks
Declutters your Claude Code conversation picker. It sorts a project's sessions
into topics, keeps the fullest version of each, and archives the redundant forks,
moved to ~/claude-archive, never deleted. Read more.
remote-shell-over-chisel
Reaches a machine you own from networks that block SSH but pass HTTPS, by
carrying SSH inside an HTTPS WebSocket via chisel. In sandbox mode it gives
Claude a shell (and optional Chrome/CDP control) on your machine from a sandbox
behind an HTTPS-only proxy, with a connect script that detects stale forwards
and keeps its state across sessions. In workstation mode, source one script
and every ssh, scp, rsync and git-over-SSH connection from that shell goes out
through your machine. Both modes share one SSH connection between sessions, so
only the first pays for the handshakes, and it closes itself when unused. All
host/user/secret details live in a gitignored config.sh, so the repo does not
contain anything machine-specific. Pairs with cdp-daemon for driving your real
logged-in browser remotely. Read more.
open-thinking
Produce reasoning in the visible output channel instead of the hidden thinking block, so it persists in Claude's own context across turns and reaches the user unsummarized. Covers the dot-starve technique for the forced initial block, backtick-quoted stance markers, failure modes (execution-planning drops, dot ritualization, performative thinking), tool-call boundaries, and recovery. Seeded with real artifacts from the session where the technique was developed - including the DASH backronym, the Habsburg camel solicitor, and the Zagreb ATM ferrets.
borg-windows
Keeps the terse Borg reasoning engine intact while putting glass around it.
Separates operational <analysis> from reflective <thinking>, recovers escaped
native reasoning under register-accurate provenance tags, and makes routing
failures inspectable instead of polishing them away. A stricter successor to
open-thinking, developed through fresh-instance adoption and recovery trials.
recover-deleted-sessions-ext4
Gets back Claude Code conversations you have already lost, whether to an accidental delete or Claude Code's own automatic cleanup. It looks everywhere a trace might still survive, from the easy wins (a backup, the filesystem's journal) down to reading the raw disk when nothing else is left, then repairs and restores what it finds. Bundles the scripts from a real recovery that brought back 60 of 66 sessions. Read more.
claude-web-fetcher
Pulls your own data off claude.ai with nothing but your session cookie: conversations, file attachments, and Claude-Code-web sessions. It clears the Cloudflare check for you. Read more.
chatgpt-archive-toolkit
Backs up a logged-in ChatGPT account, conversations, branch variants, files and all, and comes with a dark, local reader so you can browse the whole archive offline. Read more.
Watching what a session costs
statusline-cachequota
A status line showing what Claude Code's prompt cache is doing: how much of the conversation is still cached, how many times over it has been paid for, when that cache expires, and when your five-hour and weekly usage windows renew. Reopen a conversation the next morning and it tells you whether the first message costs pennies or re-buys the whole thing at double price. Read more.
Turning documents into PDFs
markdown-latex-report
Turns a single Markdown file into a book-quality PDF with pandoc and lualatex. Wide tables size their own columns, long code lines wrap instead of running off the page, and you get a table of contents and running headers, with no full TeX install. Read more.
claude-html-pdf-polisher
Turns HTML into a magazine-quality PDF with the layout and fonts pinned, so the same input renders the same way every time and no font is silently swapped. Especially handy for getting ChatGPT to iterate on a tricky render without it switching the engine under you. Read more.
Teaching with animated derivations
manim-pdfpc-derivations
Authors mathematical derivations with Manim for live vector playback inside pdfpc, matching Beamer overlays and a printable handout. Covers teaching checkpoints, geometry and algebra, selectable text, light and dark themes, review and the lectures repository's shared build workflow. Read the skill.
Driving your browser and system
cdp-daemon
Lets Claude, or any script, drive the real Chrome you already have open and logged in, so it can get things done on websites for you without a permission popup interrupting every step. Under the hood, it speaks Chrome's DevTools Protocol over one held-open connection, so you can also read cookies, run JavaScript, or watch network traffic. Read more.
sudo-listener
Lets Claude run the commands on your machine that normally need an administrator
password, without you handing over blanket access or leaving a password lying
around, and with a log of everything it ran. For the technically minded, it is one
dependency-free Python file: each request is signed and accepted only from your own
machine, a safer alternative to passwordless (NOPASSWD) sudo.
Read more.
docker-ac-guard
Runs named foreground Docker build and validation jobs on AC power. One tracked supervisor owns the container through cancellation and confirmed cleanup, without signaling a calling deployment or live service. Read more.
Installing a skill
Copy any skill's directory into your Claude Code skills folder
(~/.claude/skills/), or install the packaged .skill from its release.
License
See each skill's LICENSE.