Open source · Agent Skills
Clean Skills
Understand the work. Verify what is done.
Clear explanations and short narrated videos help people understand the work, and clean writing keeps your own notes and posts in your voice. For debate and clean rebuilds, strong models shape the contract and a fail-closed gate checks completion. Agreement between models is not evidence. The skills work in Claude Code, Codex and other Agent Skills hosts.
Get started
cargo install --locked clean-skillsclean-skills install --host claude --applyclean-skills install --host codex --apply- Released crate v0.6.1
- Claude Code and Codex
- MIT licence
- Strong models(done)
- Contract(done)
- Qualified checks(done)
- Executor(done)
- Gatewaiting for the gate…
Set up
One crate, both agents.
The 0.6.1 crate includes all six skills: API docs, debate, explanation, refactoring, video and writing. Leave off the apply flag to preview changes. The helper, the writing and video scripts and the verification kit need Python 3.9+; videos also need ffmpeg and a local voice model; text explanations need no tools.
Six skills in the collection
clean-api-docs
Finds the library version your project actually locks or installs, reads that code and the official docs for that version, stops early, and labels where each fact came from. When you allow a web search, it uses it only to find the official page; a search result never counts as evidence.
clean-debate
Several strong models work out a task independently and correct each other. The result is a qualified contract that a capable, not necessarily the strongest, model can carry out.
clean-explanation
Explains concepts and systems at the reader’s level with the smallest useful medium, from a short answer to a diagram, interactive example or narrated video. Preserves the important qualifications, checks artifacts with the available tools and reports what remains unchecked.
clean-refactoring
Rebuilds a large application from first principles instead of refactoring it in place, against acceptance checks that accept a known-good result and reject a defective one.
clean-video
Makes a short video that explains one idea, in English or Chinese, narrated by a voice model on your own machine. Writes the script for the ear, times each diagram to the line that names it, makes captions from the script, renders locally and checks every step’s frame, the loudness and the captions.
clean-writing
Cleans, fact-checks and publishes your own notes and posts, in English, Chinese or both, from text, voice transcripts, screenshots and photos. Light edits keep your words, drafts never invent your experiences, every change and correction is shown with its source, and nothing is published until you confirm.
Also in the repository
A status bar for Claude Code.
clean-status-bar is a two-row status line, separate from the crate. It shows the repository, branch and changes, then the model, effort and your 5-hour and weekly usage, coloured by pace. Once a window is over 80%, it also shows when that window resets. The installer previews unless you add --apply, changes only the status line setting, keeps an existing status line unless you pass --replace, and puts it back when you uninstall. It needs bash and jq.
curl -fsSL https://x.vibewait.ing/x/clean-skills/raw/branch/main/clean-status-bar/install.sh | bash -s -- --apply