Copy for the folder that runs 30+ live sites and 8 iOS apps from one terminal.

MIT-licensed agent skills for Claude Code and Codex. One of them grades a staged diff A-F across seven lanes, where the weakest lane caps the grade. Another argued $448.31 back out of a real Amazon account.

Suede Creator Skills is the folder of skills that runs Suede Labs, published as is. Every skill is a plain SKILL.md file for Claude Code and Codex, so you read the method before you run it, and every description carries a NOT FOR line naming the sibling that should win instead. The range is the point: the same pack grades a staged diff A-F across seven lanes and argued $448.31 back out of a real Amazon account, including a refund denied after the return window closed. It is free under MIT for the original work, with no binaries and no telemetry.

Suede evolves while you use it. Capabilities load as inspectable SKILL.md folders instead of piling into one memory page that every prompt has to carry. An installed skill costs its name and its description. The body loads only when a request matches, so adding a skill makes the pack sharper instead of heavier.

This site is the GitHub Pages surface generated from the public suede-creator-skills repo, not a separate marketing site. The page itself is the proof. It ships from the same history as the skills it describes: one author, and every change is a commit on main.

Add the marketplace. Install the pack. The folder that runs Suede Labs is on your machine in about 30 seconds.

Use suede-code to review my staged diff and give it an A-F ship grade. List the Suede skills that match this task and name the one that should lose. Run a copy audit on this page and return the exact rewrites. Build a QA checklist and stop at the first gate I cannot prove.

Find what blocks discovery and the next click, then get the exact line to replace it with.

Split major work into lanes that own separate files and report to one controller. WIP collision detection keeps two lanes off the same file, rollback trees keep a bad lane from taking the branch with it, and the handoff checklist will not close without evidence.

Use Suede Graph Flo XR when the outcome is broad, authorized, and worth maximum useful effort. It writes competing plans for the change, scores them, prunes, refutes the survivors, and builds only the winner. Light, standard, and deep cap the run at 55, 110, and 200 calls. It reads production for evidence and never deploys, and it does not expand permissions, spend authority, or model limits.

Create a local rights package before the work leaves your machine.

Create a rights package for this folder and list what is missing. Open the skill folder and read the file before you trust it, since every skill is plain Markdown. Run the package script against a real release. Pair it with the release linter and close the gaps it names.

Find release gaps before they become rights problems.

Audit this release folder and rank the gaps by what blocks the release. Read the lint rules in the repo and disagree with them before you run the lint. Open the lint script and see exactly what it reads. Create a rights package once the gaps are closed.