Each task group should have its own testing and documentation updated within the group rather than waiting to add testing and documentation

[keithboone]

On more complicated CRs which have multiple task groups, I've noted that Claude and Github Copilot will often add a Task group at the end that involves testing and documentation. This creates a failure mode where a change having 6+ task groups can fail testing at the end because something wasn't tested properly in Task Group 1 or 2, and problems cascade downward through other tasks, involving a lot of rework. My agents have a standing rule to include testing and documentation updates in each task group for the changes made during that group to ensure this doesn't happen, but this instruction really belongs in the Openspec tasks guidance given to agents.

[clay-good]

Fix opened as #1955 — the tasks artifact instruction now requires each task group to land its own tests and documentation, with a final group reserved for integration checks only.