메뉴
Bootstrap a Vers VM into a golden image with punkin-pi, Node.js, dev tools, and swarm extensions installed. Creates a committed snapshot that can be branched for self-organizing agent swarms.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/hdresearch/pi-vers --skill vers-golden-vm명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
Bootstrap a self-improving agent fleet on Vers from scratch. Walks through deploying agent-services, building a golden image, testing the loop, and running first tasks. Use when onboarding someone new to the Vers agent fleet workflow.
How Vers VM networking works — public URLs, port routing, TLS proxy, IPv6, and running services on VMs.
Contribute a bug fix or improvement to pi-v. Use when you've found and fixed an issue in the pi-v package.
SKILL.md 표시 중
SOC 직업 분류 기준
| name | vers-golden-vm |
| description | Bootstrap a Vers VM into a golden image with punkin-pi, Node.js, dev tools, and swarm extensions installed. Creates a committed snapshot that can be branched for self-organizing agent swarms. |
Bootstrap a Vers VM into a reusable golden image for punkin-pi agent swarms. The golden image includes Node.js, punkin-pi (built from source at the w/router release tag), GitHub CLI, dev tools, and any punkin packages you configure.
LLM_PROXY_KEY (sk-vers-*) for punkin agentsvers_vm_create with vcpu_count=2, mem_size_mib=2048, fs_size_mib=4096, wait_boot=true
Default 1GB disk is too small — punkin-pi source + build + packages + node need ~1.5GB.
Set the VM as active with vers_vm_use, then run the bootstrap script:
export GITHUB_TOKEN="<token>"
# Optional: override defaults
export PUNKIN_TAG="w/router"
export GIT_AUTHOR_NAME="reef-agent"
export GIT_AUTHOR_EMAIL="reef-agent@users.noreply.github.com"
bash /path/to/scripts/bootstrap.sh
The script:
/usr/local/bin/punkin to the built CLIpunkin installEdit the PACKAGES array in the script to add your own repos.
Critical check. Punkin reads packages from ~/.punkin/agent/settings.toml.
cat /root/.punkin/agent/settings.toml
# Should show: packages = [ "/opt/pi-vers", "/opt/vers-agent-services" ]
If settings.toml doesn't exist or is empty, agents will only have read/bash/edit/write — no vers_*, board_*, feed_*, etc. tools. Run punkin install <path> to fix.
The bootstrap script handles this, but verify:
tmux ls # should error "no server running"
ls /tmp/pi-rpc/ # should be empty
Stale tmux sessions or pi-rpc fifos from a previous punkin run cause golden images to fail when branched.
If you have an infra VM running agent-services, bake the connection details into /etc/environment so all processes (including punkin) inherit them. Include LLM_PROXY_KEY too if you want restored agents to inherit it automatically. .bashrc only works for interactive shells:
cat >> /etc/environment << EOF
VERS_INFRA_URL=https://<infra_vm_id>.vm.vers.sh:3000
VERS_AUTH_TOKEN=<auth_token>
LLM_PROXY_KEY=<sk-vers-key>
EOF
Replace with actual values. This ensures spawned agents can use board/feed/log/registry tools immediately.
Switch back to local (vers_vm_local) and commit:
vers_vm_commit with the VM ID
Save the returned commit_id — this is your golden image.
Update VERS_GOLDEN_COMMIT_ID in ~/.zshrc (or wherever it's set). Note: running punkin sessions won't pick up env var changes — they need a full restart.
w/router release tag)PACKAGES array, registered via punkin install → settings.toml| Variable | Default | Description |
|---|---|---|
GITHUB_TOKEN | (empty) | GitHub PAT for cloning private repos |
PUNKIN_TAG | w/router | Release tag to build punkin-pi from |
GIT_AUTHOR_NAME | reef-agent | Git commit author name |
GIT_AUTHOR_EMAIL | reef-agent@users.noreply.github.com | Git commit author email |
Punkin reads package registrations from ~/.punkin/agent/settings.toml. Always use punkin install <path> to register packages.
Golden images committed from VMs that previously ran punkin (e.g., from a lieutenant session) have stale tmux sessions and pi-rpc fifos baked in. These cause swarm spawns to fail on the branched VMs. Always build golden images from fresh VMs using the bootstrap script.
Default VM disk is 1GB. Golden images need at least 4GB (fs_size_mib=4096). After bootstrap, disk usage is ~1.5GB.
The old bootstrap used npm install -g @mariozechner/pi-coding-agent (the pi binary). The new bootstrap builds punkin-pi from source. If your swarm coordinator runs punkin, the golden image must also have punkin — otherwise the spawner will fail with punkin: command not found.