メニュー
Intelligent model load balancer for OpenClaw crons — distributes across providers by complexity and cost. Use when optimizing model selection for crons, balancing provider load, or troubleshooting model routing.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | model-orchestrator |
| description | Intelligent model load balancer for OpenClaw crons — distributes across providers by complexity and cost. Use when optimizing model selection for crons, balancing provider load, or troubleshooting model routing. |
Intelligent model load balancer for OpenClaw crons. Distributes crons across providers based on task complexity, provider health, quota status, and cost.
| Tier | Primary | Secondary | Tertiary | Emergency |
|---|---|---|---|---|
| T1 | MiniMax | Flash 3 | Kimi | (pause) |
| T2 | Flash 3 | Kimi | MiniMax | (pause) |
| T3 | Opus | Sonnet | Kimi | Flash 3 |
IMPORTANT: Not all providers use credit-based billing. Know the model before diagnosing.
| Provider | Billing Model | "Down" Diagnosis |
|---|---|---|
| Z.ai (GLM) | Rate-limit / quota system (like Anthropic/OpenAI). $360/yr Coding Max plan. NOT credits. Error 1113 ("余额不足") = quota exhausted for current period, NOT empty balance. Resets periodically. Check dashboard: https://open.bigmodel.cn/ | |
| Anthropic | Rate-limit tiers (tokens/min, requests/min). Cooldown = temporary. | |
| OpenAI | Rate-limit tiers. Monthly spend caps optional. | |
| Rate-limit per model per minute. Free tier available. | ||
| MiniMax | API credits (actual balance). | |
| Kimi | Rate-limit system. |
Rule: When a provider returns quota/limit errors, report it as "quota limit hit — resets on [date]" NOT "balance depleted / needs top-up." Check the dashboard for reset timing before recommending action.
When 2+ providers are down:
at crons to check recoverystate/cron-tiers.json — cron ID → tier mapping + metadatastate/provider-status.json — compact current provider health/quota snapshot (backward-compatible)state/provider-tracking.json — detailed provider registry: status, source, checks, quota payloads, staleness, temp-model registrystate/switches.log — audit trail of all model switchesscripts/orchestrate.sh — main orchestrator (run daily or on-demand)scripts/check-provider.sh — test a single providerscripts/scrape-quota.sh — scrape dashboards for quota infoUse when an existing UI artifact needs to be iterated to a verified 5/5 design score. Enumerate every view, run independent dead-item and design-rubric audits, apply fixes, then re-score with fresh reviewers until every view and the whole app pass. Designed for Enterprise Crew/OpenClaw operators finishing prototypes without false 5/5 claims.
Use when the operator says publish a skill, promote a skill publicly, or ship a skill to SuperAda, GitHub, and ClawHub. Sanitizes the skill bundle for personal info/secrets, publishes a GitHub source copy, updates SuperAda skill pages/install assets, and publishes to ClawHub with receipts.
Use when turning a coding goal or PRD into bounded build missions, running those missions through Codex, Droid, Cursor, or Claude Code, verifying outcomes separately, and preserving receipts. Geordi merges the former build-pipeline discipline with the installable mission runner.
Elite website image-to-code skill for Codex. For visually important web tasks, it must first generate the design image(s) itself, deeply analyze them, then implement the website to match them as closely as possible. In Codex, it must prefer large, readable, section-specific images instead of tiny compressed boards, and it must generate fresh separate images for sections or detail views rather than cropping them out of previously generated images.
GitHub repository initialization, authentication, PR governance, and cross-machine synchronization workflows.
Prepare for conferences, summits, trade shows, and investor events: research attendees/speakers/agenda, build pitch-deck briefs, schedule meetings, enrich contacts, capture session notes, and run post-event follow-ups.