← ClaudeAtlas

sanitize-skill-for-publiclisted

Use when a skill, prompt, playbook or set of notes written for private or work use is about to be published or shared (public repo, plugin marketplace, blog, team wiki), or when asked to remove names, company or product terms, ticket keys or internal links from it, anonymize it, or make it safe to open-source.
teransarathchandra/agent-skills · ★ 0 · AI & Automation · score 78
Install: claude install-skill teransarathchandra/agent-skills
# Sanitize a skill for public release A skill written during real work carries the incident it came from: a person's name, the company, a ticket key, a quote, a date. This skill turns that into a general, fictional, safe version and shows that it is clean. It prepares a copy. The owner decides whether to publish. ## Rules (non-negotiable) - **Work on a copy.** Never edit the private original in place. - **Keep the term list private.** It holds the exact words you must remove. Store it outside the folder being published. Never commit it, and never paste it into a PR, issue, commit message or online service. The scan script refuses to run if the list sits inside the scanned folder. - **Say less than you found.** In reports, give file, line and category. Show the matched word only in the local terminal, never in anything that gets posted. - **Prepare only.** Do not push, open a PR, or change repository visibility. Stop at a copy that is ready for review. - **A clean scan is not proof.** The scan finds the words you listed and common patterns. You still have to read the files. ## Steps 1. **Copy.** Copy the skill to a work folder outside the private source. Do the rest there. 2. **Build the term list.** Unless the owner already listed the private identifiers, ask how to get them, as options with one recommended: a) **Point me to the source material (tickets, notes, the original repo) and I draft the list for you to confirm (Recommended: it finds the variants people forget