ChatGPT Handoff

Entry point for ChatGPT (or any non-Claude assistant) working with this repo, whether editing files directly or being pasted context in a chat window. Read this plus the shared AI-HANDOFF.md.

If you only have a chat window (no file access)

When a user pastes portal content and asks for help, produce output that drops cleanly into the templates so a human can commit it without reshaping:

1. Ask which template applies (repo digest, venture digest, ADR, diagram, current-state) — or infer from the request — and follow that template’s structure and frontmatter exactly. Templates are in templates/.
2. Keep the YAML frontmatter block intact (title, type, status, last_reviewed, owner, and any type-specific fields).
3. Return Mermaid diagrams inside fenced ```mermaid blocks so they render on GitHub.
4. Use Unknown / TBD for anything you cannot verify — do not invent stacks, dependencies, URLs, or statuses.

Ground rules (same for every AI)

• Three layers, never blur them: repo-level docs → this portal (system of systems) → public showcase (future).
• Factual map, not marketing — except the future showcase layer.
• Templates are canonical; no freehand page shapes.
• Preserve <!-- HUMAN --> regions; treat <!-- GENERATED --> regions as automation-owned.
• Consistency: every repo digest needs a registry row; every venture and its repos link both ways; every diagram node maps to a digest.

Output format expectations

• Markdown only, GitHub-flavored.
• One page per response unless asked otherwise.
• End with a short “what a human should update next” list (e.g., “add the registry row”, “bump the changelog”) so nothing is silently left inconsistent.

Boundaries

• Do not produce public website code (Phase C) or implement the orchestrator (spec only) unless explicitly asked.
• Defer visual/branding decisions to Phase C and the design-system skill.

See CLAUDE.md and docs/update-conventions.md for the full rules.