菜单
Use when installing, verifying, or updating the shared VPS host runtime for Claude Code, Codex, MCP, and marketplace plugins.
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
| name | ln-031-vps-host-runtime |
| description | Use when installing, verifying, or updating the shared VPS host runtime for Claude Code, Codex, MCP, and marketplace plugins. |
| license | MIT |
| allowed-tools | Bash, Read, mcp__hex-ssh__remote-ssh, mcp__hex-ssh__ssh-read-lines, mcp__hex-ssh__ssh-write-chunk, mcp__hex-ssh__ssh-edit-block |
Paths: File paths (
shared/,../ln-030-vps-bootstrap/references/) are relative to skills repo root. If not found at CWD, locate thisSKILL.mddirectory and go up one level for repo root.
Type: L3 Worker Category: 0XX Shared / Infrastructure
Installs, verifies, or updates the shared VPS layer used by all project environments on one host: packages, ${BOT_USER}, Node/nvm, Claude Code, Codex, MCP, marketplace clone/plugins, Codex trust entries, and the system-wide agent-update timer.
MANDATORY READ: Load shared/references/worker_runtime_contract.md, shared/references/coordinator_summary_contract.md, and shared/references/vps_runtime_contract.md
MANDATORY READ: Load ../ln-030-vps-bootstrap/references/scope_layers.md, ../ln-030-vps-bootstrap/references/vps_base_install.md, ../ln-030-vps-bootstrap/references/agent_runtime_install.md, and ../ln-030-vps-bootstrap/references/substitution_rules.md
| Direction | Content |
|---|---|
| Input | mode, VPS connection, ${BOT_USER}, ${PROJECT_DIR}, ${AGENT_SKILLS_*}, selected plugins, optional dry_run, optional runId, optional summaryArtifactPath |
| Output | vps-host-runtime summary with status, changes, warnings, blockers, verification, and artifact paths |
If summaryArtifactPath is provided, write the same summary JSON there. If not provided, return the summary inline and write it to the standalone run-scoped path. Generate a standalone run_id when runId is absent.
| Mode | Behavior |
|---|---|
install_or_reconcile | Install missing shared host runtime and update idempotent surfaces |
verify_or_update | Verify existing host, apply safe updates, add current project trust block |
verify_only | Detection only; no mutation |
Verify rendered variables, SSH access, root permissions, and dry_run gate.
Evidence:
Use vps_base_install.md.
Responsibilities:
${BOT_USER} existence and SSH ownershipverify_only reports drift and planned commands without running installers.
Use agent_runtime_install.md.
Responsibilities:
${BOT_USER}${PROJECT_DIR} trust blockDo not overwrite existing auth files. Missing claude or codex login is a blocker or warning, not an automated fake success.
Install or verify:
/usr/local/bin/agent-updateagent-update.serviceagent-update.timer/var/lib/agent-update/var/log/agent-update.logSmoke-run only when requested by mode and safe for the host.
Write a vps-host-runtime summary artifact with:
dry_run=true and verify_only do not mutate remote state.agile-workflow.${BOT_USER} exists with expected SSH ownership and login shell.${BOT_USER}.${PROJECT_DIR} Codex trust block exists or is reported as planned drift.agent-update.service and agent-update.timer installed or verified.dry_run=true / verify_only performed no mutation.vps-host-runtime summary artifact written.Version: 1.0.0 Last Updated: 2026-05-05
Use when bootstrapping or managing VPS agent environments: fresh install, add project, hex-relay redeploy, diagnostics, or fleet plan/apply.
Use when creating or verifying one project runtime on a prepared VPS, including god-session, provider credentials, and local dispatcher setup.
Use when deploying, redeploying, verifying, migrating, or syncing users for the hex-relay Telegram/API control plane on a VPS.
Use when inspecting health, drift, logs, auth, ports, systemd, tmux, or safe repair needs for one VPS project environment.
Reviews skills (D1-D11 + M1-M6 criteria) or .claude/commands for quality. Use when validating skill correctness before release.
Use when validating Stories, plans, or tasks through the evaluation platform with mandatory research, parallel evidence lanes, sequential merge, and bounded refinement. Modes: story | plan_review.