Sync model agents and AI tooling rules

[yulonglin]

Summary

• add four OpenRouter-backed agent definitions
• add durable LLM-judge methodology rules
• sync generated Codex permission rules with current Claude permissions
• update Marp figure references from anthroplot to pastelplot
• remove the stale project-level hive disable override

Verification

• validated global and project Claude settings JSON
• preserved required statusLine, hooks, and permissions keys
• validated all agent frontmatter names and model IDs
• confirmed Codex config/rules parse with codex-cli 0.146.0
• scanned the staged diff for literal secret indicators
• ran codex-companion review; it exited 0 without findings

Intentionally excluded

• global claude/settings.json local proxy token and machine-specific generated state
• codex/config.toml machine-local marketplace paths and timestamps
• absolute /home/yulong/romp symlinks
• codex/.sandbox_migration runtime marker