本文へ移動
cccskills
無料GitHub で公開

cross-host-rig-commands

Use when addressing a registered remote OpenRig host, choosing its transport, or interpreting a cross-host result.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md5.4 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Cross-host rig commands

Use a registered host address and the entry's declared transport. SSH entries run a remote CLI through a single-hop shell; HTTP entries use the remote daemon. These paths have different prerequisites and command coverage. Check the exact command's help and the target installation before a consequential operation.

Choose and inspect the destination

rig host list shows registered hosts. rig host add --help describes registry writes; rig host doctor --help describes reachability checks. Inspect the configured OpenRig home rather than assuming the default ~/.openrig/hosts.yaml. Registry changes affect later routing, so make them within the requested scope.

hosts:
  - id: test-vm
    transport: ssh
    target: test-vm.local
    user: example-user
  - id: remote-dev
    transport: http
    url: http://remote-dev.example:7433
    bearer_env: REMOTE_RIG_TOKEN

The registry requires a hosts array, unique non-empty ids and a supported transport. SSH needs a non-empty target; user and notes are optional. HTTP needs a URL and accepts at most one of bearer_env or bearer_file. Omitting both is valid for a tokenless target. A configured but unavailable bearer is a permission failure, not anonymous fallback. Host ids also have reserved-name validation; use the actual validator's error rather than inventing an alias that collides.

Address the intended operation

rig send dev-worker@example-rig "check the assigned result" --host remote-dev --verify
rig capture dev-worker@example-rig --host remote-dev
rig ps --host remote-dev --nodes --json
rig whoami --host remote-dev

These coordination commands select SSH or HTTP from the registry entry. They do not silently try another transport when one fails. A host-qualified target can be convenient, but confirm its parsing and any persisted host selection with the particular command. Explicit --host is preferable for consequential cross-host work.

Queue destinations use explicit --host or a supported host-qualified destination; queue writes do not follow persisted host selection. The CLI separates the host from the canonical seat in the routing envelope. Different queue subcommands have different flags: check their help instead of copying a flag from send. A local success message alone does not establish remote persistence or pickup.

Read failures at the layer that failed

SignalWhat to inspect
registry-load-failed / unknown-hostRegistry path, entry and exact destination
ssh-unreachableSSH connection, target and transport diagnostics
permission-gateThe configured SSH or HTTP credentials and target policy
remote-daemon-unreachableTarget listener and actual running daemon identity
remote-outcome-unknownThe request may have arrived (timeout, dropped connection, unreadable reply): check its effect, for a send the target pane, before any retry
remote-command-not-foundRemote CLI installation and executable lookup
remote-command-failedThe remote operation's own status and error

The exact result taxonomy depends on the path. HTTP failures expose the remote status/error; SSH distinguishes the shell transport from the remote command. Do not start or replace a remote daemon merely because a diagnostic suggests it; first establish the target's state and the authority for that lifecycle action. A timed-out write may have reached its destination. Reconcile its durable effect before retrying.

Verification and attribution

Transport success is not delivery verification or agent consumption. SSH forwards --verify and returns the remote CLI result. HTTP preserves the remote transport verdict and explicitly reports that the local pane-effect check did not run cross-host. Inspect that result and the destination effect required by the task; never replace it with “SSH exited zero” or “HTTP returned successfully.”

Cross-host output names the host/target; JSON envelopes carry cross_host metadata. The precise envelope differs by command and transport. Preserve the underlying remote result when passing evidence onward.

A target's host suffix selects a destination; a sender's origin identifies where a reply belongs. Use the rendered reply address and verify its host mapping. Current send derives the sender from the executing seat context, not a caller's --from assertion. It adds origin-host identity at cross-host forwarding; do not assume local sends have the same suffix. Unknown identity remains unknown.

This guide describes the implemented command paths, not proof of a live remote journey on your installation. Forks, handovers and other lifecycle compositions need their own supported command and scoped authority; reading remote state does not authorize them.

See also

  • openrig-user — exact command reference
  • seat-continuity-and-handover — stable seat and occupant outcomes

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.

日本語の概要は準備中です。原文の説明を表示しています。

mvschwarz/openrig7,0282026年10月12日 更新

Use when designing, building, operating, or diagnosing an ongoing application whose live backend or control loop includes OpenRig agents, including applications with a Markdown, YAML, or JSON agent control plane or a thin surface over specialist agent roles.

日本語の概要は準備中です。原文の説明を表示しています。

mvschwarz/openrig7,0282026年10月12日 更新

Use when a bounded real-world procedure has a deterministic happy path but brownfield, variable, or partially knowable state; when an operation must resume from verified evidence; or when deciding whether agent judgment or ordinary code should own a procedure's control loop.

日本語の概要は準備中です。原文の説明を表示しています。

mvschwarz/openrig7,0282026年10月12日 更新

Use when creating, refreshing, packaging, inspecting, promoting, or deprecating a named per-seat starting point — Agent Starter manifest authoring, the 6-state lifecycle (captured → named → inspectable → used → promoted → deprecated), provenance honesty, and refusal rules. NOT a VM image; a managed starting point composed from agent role + startup context + optional native session source + provenance.

日本語の概要は準備中です。原文の説明を表示しています。

mvschwarz/openrig7,0282026年10月12日 更新

Use when designing or auditing how an agent becomes useful after launch — AGENTS.md overlays, role files, skills, rig specs, workflow specs, startup checklists, refocus messages, "rig context" surface. Covers the 4 failure modes that make startup context fail (old rig spec misses current operating mode; current agents never told about new guidance; startup file as dumping ground; orchestrator transmits implementation without preserving product intent).

日本語の概要は準備中です。原文の説明を表示しています。

mvschwarz/openrig7,0282026年10月12日 更新

Use before launching a team during agent-guided setup, or when a user asks to configure OpenRig command permissions, reduce repeated native approval prompts, or apply a selected rig/seat permission policy.

日本語の概要は準備中です。原文の説明を表示しています。

mvschwarz/openrig7,0282026年10月12日 更新

mvschwarz のスキルをすべて見る

このスキルの問題を報告する