Meaningful Linear issues

A good issue is not just a good description — that's the easy half. The half agents skip is the metadata that makes Linear actually work as a planning tool: estimates feed cycle velocity, labels feed breakdowns, milestones feed roadmaps, the project feeds everything. Those fields are optional in the API, so agents leave them blank and the user has to come back and ask again.

Flip the default: metadata is opt-OUT, not opt-in. Resolve and fill it before calling save_issue, in one pass — propose concrete values, don't omit.

Cardinal rule: never guess the project

The single worst failure is dropping an issue into the wrong/random project. Resolve the project in this order — stop at the first that's certain: