菜单
Free LLM query via agnes-2.5-flash (OpenAI-compatible, zero cost). Single-call text completion for agent pipelines and SelfEvolver fixes.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/bobo070314/v1.1-self-evo-factory --skill llm-query-agnes命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | llm-query-agnes |
| description | Free LLM query via agnes-2.5-flash (OpenAI-compatible, zero cost). Single-call text completion for agent pipelines and SelfEvolver fixes. |
| version | 0.2.0 |
| type | skill |
| status | live |
Free LLM access for skill pipelines. Calls https://apihub.agnes-ai.com/v1/chat/completions
with model agnes-2.5-flash (cost = 0, context 128K, max_tokens 8192).
No external dependency (standard library urllib only).
python run.py --prompt "修复这段代码" --json
python run.py --prompt "..." --max-tokens 2048 --temperature 0
python run.py --version
python run.py --help
| Arg | Description |
|---|---|
--prompt | (required) User prompt text |
--model | Default agnes-2.5-flash |
--max-tokens | Default 1024 |
--temperature | Default 0.2 |
--json | Output machine-readable JSON |
--dry-run | Print request info without calling API |
Environment variable AGNES_API_KEY (loaded from OpenClaw env). The gateway
does not verify a local key — use the real env value for external calls.
127.0.0.1:18789 — that is the Control UI port, not the LLM API.agnes-2.5-flash (no agnes/ provider prefix).Validate URLs in Markdown/HTML/text files over HTTP with status classification, retry, and JSON output.
Demo fixture for the self-healing loop: a config whose _meta.json is repaired automatically by Agnes LLM when it has syntax errors. Run scripts/test_self_healing.py to exercise the Diagnosis→Fix→Verify loop.
Industrial-grade CSS minifier: strips comments/whitespace/redundant semicolons while preserving semantics. Exit-code semantics (0/1), honest --dry-run, machine-readable JSON with savings %.
基于 SOC 职业分类