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/novotnyllc/subagent-orchestration --skill rp-reminder-v2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
基于 SOC 职业分类
| name | rp-reminder-v2 |
| description | Reminder to use RepoPrompt MCP tools |
| repoprompt_managed | true |
| repoprompt_skills_version | 61 |
| repoprompt_variant | mcp |
Continue your current workflow using RepoPrompt MCP tools instead of built-in alternatives.
| Task | Use | Not |
|---|---|---|
| Search paths/content | file_search | grep, find, Glob |
| Read file (whole or sliced) | read_file | cat, head, Read |
| Directory tree | get_file_tree | ls, find |
| Signatures / overview | get_code_structure | reading whole files |
| Edit file | apply_edits | sed, Edit |
| Create / delete / move | file_actions | touch, rm, mv, Write |
| Git status / diff / log / blame | git | shelling out for analysis |
| Tool | Use for |
|---|---|
manage_selection | Curate the file set used by chat, builder, and exports. Refresh before each planning call. Modes: full, slices, codemap_only. |
workspace_context | Snapshot current prompt + selection + token budget; also exports. |
prompt | Read/set the shared prompt; list or select copy presets. |
context_builder | Heavy discovery sub-agent — describe the task, it curates files + rewrites the prompt. response_type: clarify / plan / question / review. Pass export_response:true to hand the result to a child agent. |
ask_oracle | Chat-mode reasoning over the current selection. Continue existing chats (new_chat:false) rather than opening new ones. Modes: chat / plan / review. |
oracle_chat_log | Recover recent Oracle messages after compaction. |
ask_user | Ask the user when ambiguity is load-bearing — don't guess at requirements. |
multi_agent_v2Dispatch a sub-agent when a side investigation or delegated chunk of work would otherwise flood this session's context. Use shipped Codex agents directly, not RepoPrompt role labels. Always pass fork_turns:"none" because these agents read context from the first message. Always pass the listed reasoning_effort.
| Agent type | Reasoning | Use |
|---|---|---|
code_mapper | low | Fast read-only probes, git archaeology, wiring lookups, narrow in-repo checks. One question per probe. |
docs_researcher | low | External docs, library/API behavior, web research, standards. |
implementation_engineer | low | Well-scoped implementation work from an existing plan. |
implementation_lead | high | Complex implementation work, architecture-sensitive changes, or branch-level ownership. |
investigation_lead | high | Multi-step root cause analysis and evidence gathering. |
test_automator | medium | Focused test, benchmark, or verification work. |
browser_debugger | medium | Browser/UI runtime debugging. |
workflow_orchestrator | xhigh | Plan critique, dependency/risk analysis, coordination review. |
Key operations:
spawn_agent with task_name, agent_type, reasoning_effort, fork_turns:"none", and a self-contained message.wait_agent, then list_agents for current status.followup_task with the target task name.close_agent once you have recorded the result.Fan-out pattern: call spawn_agent once per independent probe, then use wait_agent and list_agents until every spawned agent has returned. There is no detach flag in this workflow; spawn_agent already creates independent agents.
Export handoff: when context_builder or ask_oracle returns oracle_export_path, include that path inside the child agent's next message so it reads the export with read_file.
// Search · Read · Edit · File ops
{"tool":"file_search","args":{"pattern":"keyword","mode":"auto"}}
{"tool":"read_file","args":{"path":"Root/file.swift","start_line":50,"limit":30}}
{"tool":"apply_edits","args":{"path":"Root/file.swift","search":"old","replace":"new"}}
{
Continue with your task using these tools.