CROSSTALK

CROSSTALK is a human–agent evidence canvas for safer whole-person reasoning. It uses six imperative WebMCP tools to let an agent reconstruct source-linked evidence across development, body, trauma, context and function on the same visible map a human reviews. The workflow preserves uncertainty, rejects diagnostic assertions, and blocks the final Context Passport until a human reviews and separately approves the map.

Unlike a chatbot beside a webpage, CROSSTALK lets people and agents collaborate on one inspectable interface. Agent actions create visible, auditable changes, while consequential interpretation and release remain human decisions.

Implementation: dependency-free HTML, CSS and JavaScript hosted on ChatGPT Sites. The page registers six narrow tools with document.modelContext.registerTool. Contract tests verify the tool surface, synthetic-data boundary, five-lane coverage, diagnostic-assertion rejection and human-review prerequisite.

Testing: open the live URL in ChatGPT’s in-app browser, call read_case_question and search_life_evidence, confirm create_context_passport is blocked before approval, run the 45-second case, complete human review, approve the map and create the Context Passport.

All content is fictional. This is an educational reasoning simulation, not diagnosis, treatment, legal advice, a medico-legal opinion or an accommodation assessment.