For small output changes, I would keep the marketplace Skill managed and add a separate, short team instruction or companion Skill: “After running this Skill, include a Source links column and use Chinese headings.” That lets the original installation continue to track its source when you next sync it. Test the combined result after each upstream update, because the upstream Skill may change its output structure; your extra instruction is a layer, not an automatic compatibility guarantee. See installing and managing Skills.

Make a local copy when you need to change the Skill's actual steps, scripts, permissions or evidence standard, not just presentation. Then you own that copy: upstream releases will not safely merge themselves into your edits. Keep a link to the source and its version in your copy, compare upstream changes periodically, review conflicts, and test with a real sample before updating the team. If a change would benefit everyone, contributing it upstream may reduce future maintenance.

For your example, start with the managed Skill plus a team-wide output rule, run it on one redacted client brief, and check that every row has a real source link and Chinese heading. Which part do you expect to change most often: the output format or the logic that gathers the sources?