Run Molmo live four-route agent matrix

[MiaoDX]

Summary

• replace the Molmo live cleanup matrix with four world-public-labels routes: Claude Code + MiMo TP, Claude Code + Kimi, OpenAI Agents SDK + MiMo TP, and OpenAI Agents SDK + Kimi
• make CI live report metadata carry agent_engine and use engine-specific model env vars
• update Pages assembly/index rendering and focused tests for the new live route names

Local live proof

Each route was run locally with seed=7, world-public-labels, and relocation_count=1 before this CI change. All four produced run_result.json and report.html, called MCP cleanup tools, reached done, and had semantic_acceptability.status=success with accepted_count=1/1. The one-object smoke exits nonzero under the existing five-object checker threshold, so artifact semantics were used for route-health proof.

Artifacts:

• output/local-live-proof/claude-mimo-world-public/0622_1603/seed-7/run_result.json
• output/local-live-proof/claude-kimi-world-public/0622_1607/seed-7/run_result.json
• output/local-live-proof/agents-mimo-world-public/0622_1611/seed-7/run_result.json
• output/local-live-proof/agents-kimi-world-public/0622_1614/seed-7/run_result.json

Verification

• ./scripts/dev/run_pytest_standalone.sh -q tests/unit/molmo_cleanup/test_ci_live_reports.py tests/contract/dev_tools/test_task_agent_just_recipes.py -k 'ci_live or openai_agents or provider_overrides'
• .venv/bin/ruff check roboclaws/household/ci_live_reports.py scripts/molmo_cleanup/run_ci_live_cleanup_matrix.py scripts/reports/write_pages_index.py tests/unit/molmo_cleanup/test_ci_live_reports.py tests/contract/dev_tools/test_task_agent_just_recipes.py
• .venv/bin/ruff format --check roboclaws/household/ci_live_reports.py scripts/molmo_cleanup/run_ci_live_cleanup_matrix.py scripts/reports/write_pages_index.py tests/unit/molmo_cleanup/test_ci_live_reports.py tests/contract/dev_tools/test_task_agent_just_recipes.py
• workflow YAML parse smoke for .github/workflows/ci.yml
• dry-run CI live matrix + Pages assembly smoke