メニュー
Use when creating, assigning tasks to, monitoring, or tearing down claude-worker autonomous agent VMs. Covers the full lifecycle via claude-ctl.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | claude-ctl |
| description | Use when creating, assigning tasks to, monitoring, or tearing down claude-worker autonomous agent VMs. Covers the full lifecycle via claude-ctl. |
| allowed-tools | Bash, Read |
| injectable | true |
Manage claude-worker autonomous agent VMs via the claude-ctl CLI.
initInitialise config at ~/.config/claude-ctl/config.toml. No-op if config already exists.
claude-ctl init
provision <name>Create a VM, wait for it to be ready, optionally seed a goal and stream its output.
claude-ctl provision <name>
claude-ctl provision <name> --goal "Build and deploy X"
claude-ctl provision <name> --goal "..." --watch # stream SSE output until done/failed
vmStatus=Running + IP<IP>:4200goal <name> <text>Post a goal to an already-running VM.
claude-ctl goal <name> "Refactor the auth module"
claude-ctl goal <name> "..." --watch # stream SSE until done/failed
delete <name>Tear down a VM. Idempotent — 404 is treated as success.
claude-ctl delete <name>
File: ~/.config/claude-ctl/config.toml
workspace_api_url = "https://workstations-api.sammasak.dev"
Env overrides:
| Variable | Overrides |
|---|---|
CLAUDE_CTL_WORKSPACE_API_URL | workspace_api_url |
| Field | Value |
|---|---|
| Container image | registry.sammasak.dev/agents/claude-worker:latest |
| Bootstrap secret | claude-worker-bootstrap |
| Instance type | claude-worker |
| Run strategy | Always |
| Idle halt | 1440 min (24 h) |
provision → (wait) → goal → watch SSE → delete
claude-ctl provision worker-1 --goal "..." --watch — fire and monitorclaude-ctl goal worker-1 "..."claude-ctl delete worker-1--watch streams Server-Sent Events from /goals/{id}/stream:
[DONE] — goal completed successfully[FAILED ...] — goal failed with reasontool_start, file_op, goal_loop, review_start, session_endUse when asked to document work, record decisions, create session records, write ADRs or RFCs, or update knowledge notes in ~/knowledge. Also use when confused about who a person is, what a company is, what situation the user is referring to, or any personal or professional context — read the vault before asking.
Use when creating or editing notes in an Obsidian vault, writing to ~/knowledge, or using Obsidian-specific syntax (wikilinks, callouts, embeds, frontmatter properties).
Use when reviewing any implementation. Replaces fixed spec+quality two-step with category-driven parallel specialists: an assessor picks relevant categories for the specific artifact, one specialist per category reviews in parallel, controller aggregates. Use after implementer completes a task in subagent-driven-development.
Use when dispatching work to background agents, defining custom subagents, or choosing between local agent dispatch and VM-based workers (claude-ctl).
Use when working with Kubernetes clusters, GitOps deployments, Flux reconciliation, Helm releases, or cluster troubleshooting. Guides declarative cluster management and GitOps workflows. Not for secrets encryption or SOPS operations — route those to secrets-management.
Use when sending a notification to ntfy from a homelab service, loop, or agent. Covers the canonical in-cluster publish pattern and the silent-failure trap of posting to the Authentik-gated public URL.