/expand: what the public web says you have done

The writer may use only what data/profile.json holds. This skill finds what is missing from it, from places the user already points to, and asks before adding anything.

Steps

1. Read data/profile.json: links (GitHub, portfolio, LinkedIn), the projects, the experience bullets, the skills and the facts.
2. For each public link, fetch what is there. For GitHub, the user's public repositories: name, description, languages, README highlights, stars, last activity. For a portfolio site, the project pages. LinkedIn is not fetched; the user can export it as a PDF into documents/ instead.
3. Compare with the profile. List, as a table, each candidate addition: a project not in projects, a language or framework not in skills, a concrete, checkable fact not in facts. Each row names its source. Leave out anything vague, anything older than the profile's own timeline, and anything the user did not build themselves.
4. Ask the user which rows to add. Then write them with /profile's rules: projects as {name, role, start, end, link, bullets}, skills into the right list, facts as one sentence each. Every added bullet is plain, has one fact, and uses only what the source said.
5. Say that the answer memory starts over, because the profile changed, and offer npx jev apply --dry --count 2 to see the effect.

Never invent a number, a date or a credential. If a source and the profile disagree, ask; do not pick.