メニュー
Recover Satelle failures. Use when setup, run, steer, status, logs, MCP, provider auth, readiness, transport, Host Daemon, or session operations fail or produce blockers.
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
Setup Satelle hosts from an AI-agent workflow. Use when configuring a host, planning onboarding, installing or updating the Host Daemon, selecting provider auth, validating readiness, or preparing MCP/client integrations.
Route Satelle work to the narrow setup, use, or recover skill. Use when an agent is asked to configure, operate, troubleshoot, or explain Satelle but the correct Satelle workflow is not obvious.
Operate Satelle sessions. Use when running prompts, steering existing sessions, checking status, reading logs, following events, stopping turns, choosing output formats, or continuing remote computer-use work.
| name | satelle-recover |
| description | Recover Satelle failures. Use when setup, run, steer, status, logs, MCP, provider auth, readiness, transport, Host Daemon, or session operations fail or produce blockers. |
Recover from evidence. Satelle failures should end with a typed error, preserved state, and a narrow next command.
Rerun the failing command with the command's --json mode. Use --error-format json only when the current binary exposes that global option. Preserve the command, host, profile, session id, exit code, and error code.
Completion criterion: the agent can name the stable error code and the affected target.
Run satelle config check --json and satelle config explain --json for the selected context.
Completion criterion: local config is valid, or config errors are fixed before remote diagnosis continues.
Run satelle doctor --json, narrowed with --scope when the failure already points to transport, config, codex, computer-use, or provider.
Completion criterion: every blocker has a scope, finding, evidence, and recovery command.
Use satelle logs --json, scoped by --session when a session id is known. Do not request raw diagnostic exports unless the user explicitly accepts the sensitivity.
When available, prefer redacted task artifacts such as plan.md, worklog.md, and goal.md alongside normalized logs before asking for raw transcripts, screenshots, or recordings.
Completion criterion: claims about prior actions are backed by normalized log entries, current status, or redacted task artifacts.
Use satelle setup --dry-run, satelle repair --dry-run, or satelle host update --dry-run before applying changes. Ask before using --yes.
Completion criterion: the recovery plan identifies what will mutate, what state is preserved, and how to verify afterward.
Prefer redacted diagnostic bundles when the current binary exposes them. Avoid raw exports, screenshots, recordings, full transcripts, and provider payloads unless the user explicitly requests them.
Completion criterion: the artifact scope and sensitivity are clear before capture.
