菜单
Configure and use named pi model/tool/thinking presets for planning, execution, and subagents. Use when selecting or standardizing models by task role.
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
Plan and execute work with dedicated planner, executor, and subagent flows in pi. Use for multi-step tasks, delegation, verification, or when switching models/tools by role.
Open GitHub pull requests with newline-safe bodies. Use when asked to create, open, draft, or update a PR description/body.
Use parent-controlled WezTerm Matrix panes for visible multi-agent orchestration. Use when the user asks for Matrix, WezTerm panes/tabs/splits, visible subagents, or multi-agent cockpit workflows.
Review GitHub PRs or SourceHut patchsets with forge CLI tools, local checks, and concise actionable findings. Use when asked to review a PR, patch, branch, or merge request.
Triage GitHub Actions or SourceHut CI/build failures. Use when asked why CI is failing, what checks ran, or how to fix a failed build.
Use pi's read-only plan mode with elicitation and isolated git worktree execution. Use when a task should be planned before changes, needs user approval, or should run safely outside the active checkout.
| name | model-presets |
| description | Configure and use named pi model/tool/thinking presets for planning, execution, and subagents. Use when selecting or standardizing models by task role. |
This repo includes an agent-presets extension that reads .pi/agent-presets.json.
Copy the example and edit model ids for your account/provider setup:
cp .pi/agent-presets.example.json .pi/agent-presets.json
{
"presets": {
"planner": {
"provider": "anthropic",
"model": "claude-opus-4-1",
"thinkingLevel": "high",
"tools": ["read", "bash", "mcporter_list", "elicit_plan_questions", "create_git_worktree"]
}
}
}
model may also be provider/model-id when provider is omitted.
/preset — list presets/preset planner — apply one preset/plan ... — read-only plan mode plus prompt/execute-plan — execute the last captured plan/execute-plan worktree <name> — execute the last captured plan in an isolated worktreerun_subagents — run parallel pi -p subagents; jobs default to current/heuristic model selection, set each job's model for multi-model comparison