菜单
Configure a side-by-side codex-glm command that routes Codex CLI to Huawei Cloud MaaS glm-5.1 through a CCR /v1/responses shim while preserving the original codex command.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Deploy, verify, upgrade, or roll back the anthropic_stream_guard LiteLLM plugin on a docker-compose LiteLLM gateway. Use when asked to install/uninstall the plugin, wire a LiteLLM deployment for native Claude Code clients, or verify the gateway stream fix.
Install, configure, verify, or troubleshoot a side-by-side `codex-forky` command that routes normal Codex tool/code-execution turns through an existing forky service to the MaaS execution backend such as LiteLLM/Huawei MaaS GLM, while routing non-tool ordinary turns and image turns directly to Codex ChatGPT/OAuth. Assumes forky is already installed and running, usually by claude-code-maas-hybrid-router.
Nightly (or on-demand) memory hygiene for AI coding agents — the thing native auto-memory does NOT do. Runs a "dream" pass over project memory to deduplicate entries, validate that every referenced file path still exists (so the agent never acts on "confidently wrong" stale references), compress survivors into a compact index, and detect when a new finding conflicts with an approved rule or CLAUDE.md — emitting a review-required conflict candidate and a proposed CLAUDE.md patch (never auto-applied). Use when memory has grown noisy/stale, on a nightly schedule, or whenever the user asks to dream/consolidate/clean up project memory, distill repeated workflows, or recall approved context.
Install and configure forky (github.com/vladharl/forky) end-to-end so `claude-forky` routes Claude Code planning, image, and classifier traffic through Claude OAuth while sending normal tool/code execution to Huawei Cloud MaaS (GLM-5.2) through an existing local LiteLLM proxy. Use when the user wants one Claude Code command with Claude-quality reasoning and vision plus lower-cost MaaS execution, while preserving plain `claude` for Claude.ai OAuth connectors. Sets up forky as a systemd user service, applies local compatibility patches (vision routing, Claude Code system/developer message-role normalization, and Anthropic cache-control TTL ordering), writes a `claude-forky` wrapper that uses ANTHROPIC_BASE_URL without ANTHROPIC_AUTH_TOKEN, adds copy-friendly mouse settings, trusts `/root` when requested so local permissions are honored, and merges plan-mode hooks into `~/.claude/settings.json`. Assumes LiteLLM is already running on :4000 and that the user has logged into Claude Code (`claude /login`) so OAuth c
Deploy, configure, validate, troubleshoot, or extend an OpenAI-compatible API proxy backed by PostgreSQL, Prometheus, and Grafana, routing through Huawei ModelArts MaaS (ap-southeast-1) with multi-key load balancing. TRIGGER when the task involves LiteLLM proxy deployment, Docker Compose stack with litellm_config.yaml, Huawei MaaS model routing, virtual key or budget management, Prometheus/Grafana observability for LLM traffic, custom_callbacks.py TTFT/TPOT/ITL metrics, multi-key load balancing, an optional self-hosted SearXNG search MCP (`web_search`/`fetch_url`) as an alternative to Exa, wiring a `claude-glm` (claude-code-router) client through the proxy, or any reference to `LITELLM_MASTER_KEY`, `HUAWEI_MAAS_API_KEY`, `MCP_TOKEN`, or `docker compose` with this stack.
Execute and troubleshoot Huawei Cloud server migration in this repository. Support both current batch existing-target SMS workflow (`terraform apply` + `scripts/mgc_sms_existing_target_batch.py`) and single-source SMS/rsync workflow (`scripts/mgc_migrate.py`). Use when users ask to migrate ECS/on-prem VMware, run migration in `ap-southeast-3`/`la-*`, resume after pause from `out/resume_checkpoint_latest.json`, monitor tasks in `out/task_poll_latest.json`, or diagnose SMS/MGC errors such as SMS.1404, SMS.0515, SMS.6504, SMS.6602, SMS.6603, SMS.6617, SMS.7605, SMS.7703, and SMS.8115. Also use for Chinese requests like “跨区域迁移”, “MGC/SMS 迁移流程”, “批量迁移恢复”, or “迁移排障”.
| name | codex-huawei-maas |
| description | Configure a side-by-side codex-glm command that routes Codex CLI to Huawei Cloud MaaS glm-5.1 through a CCR /v1/responses shim while preserving the original codex command. |
Use this skill when the user wants codex-glm to use Huawei MaaS glm-5.1 without changing the original codex command.
export API_KEY='replace-with-your-maas-api-key'
./scripts/configure-codex-glm.sh
./scripts/test-codex-glm.sh
codex-glm exec --skip-git-repo-check --ephemeral "Reply with OK only"
~/.codex/glm.config.toml and ~/.codex-glm/model-catalog.json and launches Codex with --profile glm.codex command untouched./chat/completions.POST /v1/responses.~/.config/codex-glm/env with 0600 permissions.glm-5.1, context window 120000, and max output tokens 8192.glm-5.1 metadata without the fallback warning.CODEX_GLM_ENABLE_SEARCH=1 routes default traffic through LiteLLM /v1/responses and installs the local search transformer.CODEX_GLM_ENABLE_IMAGE=1 advertises image input and adds a CCR image route through LiteLLM /v1/chat/completions.CODEX_GLM_IMAGE_MODEL selects the LiteLLM model group for image requests and defaults to vision-openrouter.custom_callbacks.py search/image hooks used by claude-glm.RESTORE_CCR=1 ./scripts/configure-codex-glm.sh
ccr restart
The v1 route shim supports text and the practical Codex coding-agent tool subset. Image input and web search are disabled by default and require the LiteLLM opt-in routes above. File search, remote MCP passthrough, and full Responses item streaming are still out of scope.