菜单
Use when the user wants to start an AI agent (Claude, Codex, Gemini, Aider) on a remote workspace server for a project. Use this even if they just say "launch claude on myproject", "start an agent", "open a session on the server", "connect to workspace", or "I want to work on project X remotely".
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
Use when the user wants to dispatch a coding task to a remote workspace — fixing bugs, implementing features, running an agentic coding run, or creating a PR automatically. Use this even if the user just says "fix the login bug on myproject" or "run this task" or "create a run" or "implement X on Y". Also use when the user asks to approve or reject a pending run.
Full development workflow orchestration — requirements analysis, architecture planning, implementation, code review, and testing using specialized sub-agents with approval gates. Use when the user wants to develop a complete feature end-to-end, or asks for the full development flow. Triggers on /dev-flow.
Use when the user wants to manage projects on the platform — listing configured projects, adding a new repo, updating project settings, or checking if a project is cloned on its workspace. Use this for "show my projects", "add this repo", "register a new project", "which projects do I have", "is this project ready on the workspace", or any project configuration task.
Use when the user asks about workspace servers — listing them, testing SSH connectivity, checking what agents are installed, viewing server details, or troubleshooting server issues. Use this for "show me my servers", "is the server reachable", "what agents are on server X", "test connection to workspace", or any infrastructure-level question about the remote machines.
Use when the user wants to interact with a running agent session on a workspace server — sending commands, reading output, listing active sessions, or closing them. Use this for "send this to the agent", "what's the agent doing", "check session output", "close that session", "list my sessions", or any tmux-based agent interaction.
Use when the user asks about what's happening on the platform — active runs, failed tasks, pending approvals, system health, or analytics. Use this for "what's running", "any failures", "show me the status", "is everything healthy", "how many runs today", or any question about current platform state.
| name | autodev-launch |
| description | Use when the user wants to start an AI agent (Claude, Codex, Gemini, Aider) on a remote workspace server for a project. Use this even if they just say "launch claude on myproject", "start an agent", "open a session on the server", "connect to workspace", or "I want to work on project X remotely". |
Launch an AI agent on a project's workspace server with automatic readiness checking.
API Base: http://localhost:8000/api
/autodev-launch — Interactive: list projects, pick one, check workspaces, launch
/autodev-launch <project-slug> — Launch on specific project
/autodev-launch <project-slug> <agent> — Launch specific agent (default: claude)
If no project specified, list and ask:
curl -s http://localhost:8000/api/projects | python3 -c "
import json, sys
for p in json.load(sys.stdin):
ws = len(p.get('workspace_server_ids', []))
if ws > 0:
print(f\" {p['project_slug']:30s} {p['repo_owner']}/{p['repo_name']} ({ws} workspace{'s' if ws != 1 else ''})\")"
Only show projects that have workspace servers assigned.
curl -s http://localhost:8000/api/projects/<PROJECT_ID>/workspace-readiness | python3 -c "
import json, sys
data = json.load(sys.stdin)
for w in data['workspaces']:
icon = '✓' if w['status'] == 'ready' else '✗'
print(f\" {icon} {w['server_name']:20s} {w['status']:12s} {w.get('path', '')}\")"
not_cloned or unreachable, ask if user wants to create a run first (which will clone the repo via workspace_setup phase)curl -s -X POST http://localhost:8000/api/sessions \
-H "Content-Type: application/json" \
-d '{
"workspace_server_id": <SERVER_ID>,
"agent_name": "<AGENT>",
"project_id": "<PROJECT_ID>",
"workspace_path": "<PATH>",
"user_context": "root",
"display_name": "<AGENT> @ <PROJECT_SLUG>"
}'
Use the path from the readiness check as workspace_path.
Default agent is claude. Default user_context is root.
Extract session details and report to user:
curl -s http://localhost:8000/api/sessions/<SESSION_ID> | python3 -c "
import json, sys
s = json.load(sys.stdin)
print(f'Session created:')
print(f' ID: {s[\"id\"]}')
print(f' UUID: {s[\"session_id\"]}')
print(f' Agent: {s[\"agent_name\"]}')
print(f' Server: {s.get(\"server_name\", \"unknown\")}')
print(f' Status: {s[\"status\"]}')
print(f' Tmux: {s[\"tmux_session\"]}')
print(f' Path: {s.get(\"workspace_path\", \"\")}')
"
Tell the user:
/autodev-session send <session-id> <message> to send commands