Apply Superpowers benchmark gates

[devRonPark]

Summary

• Add Tasks 4.14-4.19 for the Superpowers benchmark application work.
• Align CLAUDE, AGENTS, README, and BLUEPRINT around spec -> plan -> isolated work -> TDD -> fresh verification -> review -> finish.
• Strengthen task decomposition, test-agent, review, git helper, and RUN_REPORT evidence rules.
• Add task-local .harness state and planning artifacts for the completed work.

Verification

• PASS: grep -q 'fresh verification' CLAUDE.md && grep -q 'TDD' README.md
• PASS: grep -q 'Files' agents/task-decomposer.md && grep -q 'Verification' agents/task-decomposer.md
• PASS: grep -q 'fresh verification' agents/test-agent.md && grep -q 'TDD' agents/quality-gates.md
• PASS: grep -q 'Spec compliance' .agents/skills/harness-review/SKILL.md && grep -q 'Code quality' .agents/skills/harness-review/SKILL.md
• PASS: grep -q 'worktree' .agents/skills/branch-checkout/SKILL.md && grep -q 'verification' .agents/skills/pr-create/SKILL.md
• PASS: grep -q 'TDD' templates/skeleton/.harness/RUN_REPORT.md && grep -q 'Verification' templates/skeleton/.harness/RUN_REPORT.md
• PASS: python3 scripts/validate_tasks.py
• PASS: python3 scripts/sync_plans.py --check

Acceptance evidence

• Tasks 4.14-4.19 are marked done in tasks/index.json with matching RUN_REPORT evidence under .harness/tasks/.
• Each task RUN_REPORT records TDD red-to-pass evidence or explicit evidence status, fresh verification, and inline review approval.

Residual risk

• Documentation-only workflow changes; no runtime stack tests apply.
• PR is draft for final human review before merge.