Context format reference

A meeting becomes a bundle: a list of context items, each with a type, a title, a body, and the evidence it rests on.

• Types: decision, requirement, constraint, risk, open_question, glossary, action_item.
• Evidence: one or more quotes, each with the speaker and the timestamp it was said at.
• Status: verified means every quote in the item was found verbatim in the transcript (tolerating case, whitespace, and punctuation). Anything that fails is kept as unverified, with the reason, rather than silently dropped — but only verified items reach a ticket, a search result, or an assistant's prompt.

Fetching it

• GET /v1/bundles/:id/markdown — the bundle as markdown, quote under every claim.
• GET /v1/tickets/:key/context — the same context, addressed by the Jira key it produced.
• GET /v1/context/search?q= — search verified items by what was said, not only by title.
• The MCP tools on the Cursor & Claude page wrap the same data for an assistant, rather than a person, to read.