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/Q00/ouroboros --skill ouroboros-config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 | ouroboros-config |
| description | Open or drive the Ouroboros settings GUI (browser, TUI, or conversational fallback) |
Settings for ~/.ouroboros/config.yaml: per-stage runtime/model selects,
global runtime + LLM backend, install badges for missing CLIs, and
env-override warnings.
ooo config
/ouroboros:ouroboros-config
Trigger keywords: "ooo config", "open settings", "configure ouroboros", "change model", "change agent"
Pick the branch that matches where you (the agent) are running. The decisive question: can the user open a browser pointed at this machine?
Launch in the background (the command serves until stopped):
if command -v ouroboros >/dev/null 2>&1; then
ouroboros config
else
uvx --python '>=3.12' --from 'ouroboros-ai[tui]' ouroboros config
fi
The command detects the non-interactive context itself and serves the
settings app over a local web server, auto-opening the user's browser. The
uvx fallback is required for a Marketplace-plugin-only install, where the
MCP server exists but ouroboros is not on PATH. In a development
checkout use uv run ouroboros config.
Relay the http://localhost:<port> line from the output so the user can
open it manually if the browser did not pop up.
Tell the user: edit → Save → then ask you to stop the server. Remind them
a running MCP server may need a reconnect to pick up backend changes. Tell
them they can reopen these settings any time with ooo config; saving a
model choice never locks it permanently.
The user cannot see a browser opened here, but may be able to reach this host. Serve without auto-open and hand over the URL:
ouroboros config --web --host 0.0.0.0 --no-browser
Relay the printed URL with this host's address substituted, plus the SSH
tunnel fallback the command prints
(ssh -L <port>:localhost:<port> <this-host>).
Do NOT start a server nobody can reach. Drive the same settings conversationally over the scriptable surface:
Show the current state:
ouroboros config show
Present the user a short menu in chat — default agent, per-stage agents, per-stage models — with the current values, and ask what to change.
Apply each choice with the validated setter (same write path as the GUI):
ouroboros config set orchestrator.runtime_backend <agent>
ouroboros config set orchestrator.runtime_profile.stages.<interview|execute|evaluate|reflect> <agent>
ouroboros config set clarification.default_model <model> # interview & seed
ouroboros config set execution.default_model <model> # execute
ouroboros config set evaluation.semantic_model <model> # evaluate
ouroboros config set resilience.reflect_model <model> # reflect
ouroboros config set llm.backend <backend> # internal LLM calls
Confirm with ouroboros config show and summarize what changed.
If a set is rejected, relay the validation error verbatim — it lists the
valid keys/values.
If the command fails with a missing-dependency hint, relay it verbatim
(pip install 'ouroboros-ai[tui]'). Scriptable edits always remain on
ouroboros config show|set|backend|init|validate.
End your final message with the state breadcrumb footer (RFC #1392), e.g.:
◆ Settings GUI serving at <url> → next: Save in browser, then stop the server
◆ Config updated via chat (<keys>) → next: reconnect MCP if the backend changed
Your final response MUST end with exactly one breadcrumb footer line:
◆ <current state> → next: <recommended action>
Derive <current state> from live session state via ouroboros_session_status when that MCP projection is available; otherwise derive it from this skill's actual outcome. Never use a linear Step N of M footer because Ouroboros is an evolutionary loop. When the next action is genuinely a choice, list 2-3 honest options in the next: clause. The breadcrumb line must be the last line of the response.
基于 SOC 职业分类