Release v7.122.0

Summary

Skill descriptions now hold only trigger hints, so agents load a skill only when the request matches it, and review loops stop at a fixed round cap.

Changed

• Every Cratis skill and bundled worker skill description now says only when to use the skill and when not to, with no summary of its contents (part of #948)
• The skill authoring rule requires trigger-only descriptions, and corpus verification rejects descriptions that summarize (part of #948)
• The review-rounds skill and review report rule bound review loops to a hard round cap, with delta rounds that check earlier findings; adapted from tstack under the MIT license (part of #948)