Release v7.51.0

Added

• An authoring rule for skills, rules, agents and prompts that loads when one of them is being written or reviewed: what a skill is, how a description routes (what it does, when to use it, what it is not for), size targets and progressive disclosure, timeless text, instruction style, one source of truth, MCP tool naming, scripts for deterministic steps, evaluations, what belongs in always-on files, and a review checklist. Harnesses without path-scoped rules find it through general.md. (part of #674)
• Verification checks that every skill description says when to use the skill and is written without "I" or "you", that every file in a skill is reachable from its SKILL.md, that general.md stays within 200 lines, and that every skill either has evaluations or is listed as unevaluated with a reason; new skills cannot be added to that list. (part of #674)
• Warnings, ahead of a later switch to failures, for descriptions without a "Not for" boundary or with process words, skills close to the size cap, time words, versions and issue references in rules, agents and prompts, and provenance pages in skills. (part of #674)

Changed

• The README's skill-authoring guidance and the skill review prompt follow the new rule; skills no longer carry version pins. (part of #674)