v9.37.0

Added

• Add provider-aware prompt-size preflight with summarize, truncate, and fail strategies plus oversize run telemetry for multi-provider dispatch.
• Add per-agent status ledgers and visible agent summary tables so multi-LLM workflows show ok, degraded, failed, and timeout providers before synthesis.
• Add research breadth routing for light, standard, and exhaustive fanout with status-aware synthesis attribution.

Changed

• Strengthen /octo:research and Discover guidance to build dynamic multi-provider fleets across Codex, Gemini, Copilot, Qwen, OpenCode, Ollama, Perplexity, OpenRouter, Cursor Agent, and Claude.
• Promote named option and comparison prompts to debate so substantial "A or B" decisions route through multi-model scoring instead of plain chat.
• Regenerate Claude, Codex, OpenClaw, and Factory surfaces, including the generated octo-discipline command.

Fixed

• Route setup/configure aliases and mistyped /octo:* commands to canonical commands with fuzzy suggestions.
• Skip failed or rejected provider outputs during aggregation while preserving visible failure reasons in summaries.
• Surface oversize provider rejections instead of allowing empty outputs to look like successful provider contributions.