zai-glm53-fleetlisted

# GLM-5.3 fleet Official (2026-08-15): GLM-5.3 is live on the **Coding Plan**. The metered API is still “coming soon.” Requests for `glm-5.2` / `glm-5.1` on the coding host auto-route to 5.3. ```bash bin/zai-glm53 doctor --json bin/zai-glm53 install bin/zai-glm53 system --json bin/zai-glm53 cyber "security audit" --json bin/zai-glm53 probe --json bin/zai-glm53 budget --json ``` ## Hard rules | Never | Always | |-------|--------| | Default to `https://api.z.ai/api/paas/v4` | `https://api.z.ai/api/coding/paas/v4` | | Count Coding Plan tokens against the $10 cap | Cap only metered API + OpenRouter `z-ai/glm-5.3` | | Send `thinking.type=disabled` to 5.3 | `thinking.type=enabled` + `reasoning_effort` low\|high\|max | | Print `Z_AI_API_KEY` | Key stays in `~/.hermes/.env` / Keychain | | Hijack Claude Code off Anthropic | Write `~/.hermes/glm53-claude.env`; mutate settings only if already on z.ai | ## Smart effort - **low** — summaries, routing, routine - **high** — code / fix / implement / debug (default for agent work) - **max** — architecture, security, audit, cyber ## Harnesses - Hermes `zai-coding-glm` / `zai-coding-glm53` → coding URL, model `glm-5.3` - `hermes-yolo` → `HERMES_YOLO_MODEL=glm-coding` via LiteLLM `:4010` (5.2 id auto-routes) - Metered alias `zai-glm53-metered` is budget-gated, never default - Claude / Cline / OpenCode: source `~/.hermes/glm53-claude.env` (`glm-5.3[1m]`) - System-wide persist: LaunchAgent `com.igor.zai-glm53-env`, `~/.zshrc` marked block,