Align Case eval criteria across live and SDK authoring loops

[tmatup]

Several Maestro Case tasks in the 2026-08-30 same-ground run grade live-skill implementation details instead of the requested artifact semantics:

• optional stages/tasks are rejected when isRequired is omitted, although omission is the SDK canonical/default false representation
• HITL result-downstream requires tasks/tasks.md even when the SDK writes an equivalent root tasks.md and the compiled caseplan contains the required binding
• the SDD-only registry handoff checker accepts only the live skill's top-level list, although both arms emitted semantically complete audit envelopes with different keys
• the logging task asks for the live skill's private journal contract without stating that output contract in the prompt, leaving the SDK arm to invent a different issue-log format
• project/solution discovery rejects a substantive Case project when a generated trigger-only duplicate scaffold is also present

Update prompts/checkers so both authoring loops receive the same observable requirements and equivalent representations are accepted. Preserve negative controls for genuinely missing optionality, binding, resource identity, issue-log evidence, and distinct project ambiguity.

Evidence: exact cohort from adhoc-2026-08-30_04-43-27, with fresh matched reruns under SDK 3.27.6 and gpt-5.6-luna.

🤖 Generated with Claude Code
Co-Authored-By: Claude