메뉴
Call a configured LLM model directly through the local script using provider settings from config.yaml. Use this skill when the user wants a raw model call, prompt test, provider/model comparison, or asks to send text to a specific GPT/Claude/Gemini model. Do not use it for normal Mavis agent execution.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Debug why a session/agent/daemon behaved incorrectly. Load when user mentions a session id (ses_/mvs_*), wants logs, root-cause analysis, or asks about stuck runs, retries, permissions, or recovery. Keywords: 排查, 调试, 卡住, 为什么, log, debug, inspect, retry, recovery.
Coordinate a Mavis multi-agent team plan. Use only when the user explicitly invokes /mavis-team or /team, or 100% unambiguously asks to use an agent team / multi-agent team. Do not infer team use from complexity, deep research, long-running work, parallelism, specialist value, or verification risk.
Nightly batch skill maintenance. Reads pending signals and proposals, plans which skills to create/refine via team engine. Managed by daemon's internal scheduler — not visible to agents or users. Do not load this skill manually -- it is designed for automated nightly runs.
Create one agent on disk. Load when you need to add a new role to the team — typically called by `mavis-team` (when planning analysis says no existing agent fits and the user has consented), by `init` (when bootstrapping `.harness/reins/`), or directly when the user says 'add an agent for X' / 'new an agent' / '加一个 agent' / '建一个 rein'. Two output paths: `~/.mavis/agents/<name>/` (default — cross-project helper) or `<repo>/.harness/reins/<name>/` (when caller specifies project target — coding project rein). Do NOT load to decide WHETHER to create (that lives in `mavis-team` router) or to create a skill (use `skill-creator`).
Coordinate a Mavis multi-agent team plan. Use only when the user explicitly unambiguously asks to use an agent team
Unified DOCX skill — create, template-apply, edit/fill, read, repair, and compare Word documents. Use for formal Word deliverables and DOCX diagnosis. Not for PDF/PPT or casual plain-text drafting.
| name | llm-call |
| description | Call a configured LLM model directly through the local script using provider settings from config.yaml. Use this skill when the user wants a raw model call, prompt test, provider/model comparison, or asks to send text to a specific GPT/Claude/Gemini model. Do not use it for normal Mavis agent execution. |
| descriptions | {"zh-Hans":"直接调用配置好的 LLM 模型，用于原始模型调用、prompt 测试和 provider/model 对比。"} |
Replace <skill_dir> with the actual skill path shown by the loader.
--model provider/model. If the user didn't name a specific model, pick a sensible default or run --list first to check available models.--system, --max-tokens, --temperature, --stream, or --config only when the task clearly requires them.~/.mavis/config.yaml. Use --config when calling a non-default profile explicitly.@ai-sdk/anthropic -> messages@ai-sdk/openai -> chat/completions@ai-sdk/google -> models/{model}:generateContentThe script is a plain .py file — pick the Python launcher that exists on the host:
| Platform | Launcher |
|---|---|
| macOS / Linux | python3 (preferred) or python if it points at Python 3 |
| Windows | py -3 (preferred) or python |
Example invocations (substitute the launcher above for <py>):
<py> <skill_dir>/scripts/llm_call.py --model anthropic/claude-sonnet-4-6 --prompt "Explain this in one sentence"
<py> <skill_dir>/scripts/llm_call.py --model gemini/gemini-2.5-pro --system "Be brief" --prompt "Summarize this"
<py> <skill_dir>/scripts/llm_call.py --model minimax-test/MiniMax-M3 --timeout 600 --stream --prompt "Long planning task"
<py> <skill_dir>/scripts/llm_call.py --list
Do not assume python3 exists on Windows — it is not part of a default install. Use py -3 or
the launcher resolved at runtime.