Claude Handoff

Entry point for Claude / Claude Code working in this repo. Read this plus the root CLAUDE.md (the authoritative maintenance contract) and the shared AI-HANDOFF.md.

Operating context

• The primary maintenance rules live in CLAUDE.md — treat it as binding.
• This repo is markdown-first; there is no build/test step in Phase A. “Verification” means internal consistency, not running code.
• The .claude/skills library is available. Relevant skills for this portal: site-architecture, documentation-writer, senior-architect, design-system, ui-design-system, landing-page-generator (Phase C), knowledge-ops (hygiene), create-llms / update-llms (index), dependency-auditor (dependency facts).

How Claude should work here

1. Plan against the templates. Every new page maps to a file in templates/.
2. Make consistent edits atomically. When you add a repo, edit the digest, the registry, the owning venture, any affected diagram, and the changelog in the same change — not piecemeal.
3. Use explicit placeholders (> **TODO:**, <!-- PLACEHOLDER -->, Unknown) — never silent gaps or invented facts.
4. Preserve <!-- HUMAN --> regions; only touch <!-- GENERATED --> regions when acting as (or standing in for) the orchestrator.
5. Verify before finishing against the Definition of Done in CLAUDE.md: consistency, live diagrams, updated current-state, ADR if a decision was made.

Boundaries

• Do not build the public showcase (Phase C) unless explicitly asked.
• Do not implement the portfolio-portal-orchestrator — it is a spec only (skills/portfolio-portal-orchestrator/SPEC.md). When maintaining pages by hand, mirror what that skill will produce so future automation is a drop-in.

Handing off to another agent

Leave the repo internally consistent and note any open threads as > **TODO:** lines and/or a changelog entry so ChatGPT or a future Claude session can pick up cleanly.