delegate

Invoke the anymodel:anymodel-runner subagent via the Agent tool (subagent_type: "anymodel:anymodel-runner"), forwarding the raw user request as the prompt. anymodel:anymodel-runner is a subagent, not a skill — do not call Skill(anymodel:anymodel-runner) (no such skill) or Skill(anymodel:delegate) (that re-enters this command and hangs the session). The command runs inline so the Agent tool stays in scope; forked general-purpose subagents do not expose it. The final user-visible response must be the engine's output verbatim.

Raw user request: $ARGUMENTS

Flag reference (these are runtime-selection flags — preserve them for the forwarded delegate call, do not treat them as part of the natural-language task text):

• --engine codex|claude — selects the executor harness. codex runs the task through codex app-server; claude runs it through claude -p --output-format stream-json. Default is the configured default engine (see /anymodel:setup).
• --model <provider/model> — selects what the harness thinks with. Provider prefixes supported by the registry: zai/ (Z.AI GLM models), ollama/ (Ollama Cloud gpt-oss / qwen3-coder / minimax), opencode-go/ (glm / kimi / qwen / deepseek). When omitted, the engine's default model is used. Aliases (e.g. spark) resolve via config.
• --bridge builtin|litellm — selects how non-native providers are translated to the engine's wire format. builtin uses the built-in shim (no external dependency); litellm proxies through a LiteLLM bridge instance. Default is builtin.
• --write — grants the engine write access to the workspace sandbox (read-only is the default unless the engine has a real sandbox and --write is passed). Never default --write on engines without real sandboxing.

Execution mode:

• If the request includes --background, run the anymodel:anymodel-runner subagent in the background.
• If the request includes --wait, run the anymodel:anymodel-runner subagent in the foreground.
• If neither flag is present, default to foreground.
• --background and --wait are execution flags for Claude Code. Do not forward them as part of the natural-language task text.
• If the request includes --resume, do not ask whether to continue. The user already chose.
• If the request includes --fresh, do not ask whether to continue. The user already chose.
• Otherwise, before starting the engine, check for a resumable thread from this Claude session by running:

node "${CLAUDE_PLUGIN_ROOT}/scripts/companion.mjs" delegate --resume-candidate --json

• If that helper reports available: true, use AskUserQuestion exactly once to ask whether to continue the current engine thread or start a new one.
• The two choices must be:
  • Continue current engine thread
  • Start a new engine thread
• If the user is clearly giving a follow-up instruction such as "continue", "keep going", "resume", "apply the top fix", or "dig deeper", put Continue current engine thread (Recommended) first.
• Otherwise put Start a new engine thread (Recommended) first.
• If the user chooses continue, add --resume before routing to the subagent.
• If the user chooses a new thread, add --fresh before routing to the subagent.
• If the helper reports available: false, do not ask. Route normally.

Operating rules:

• The subagent is a thin forwarder only. It makes exactly one Bash call to invoke node "${CLAUDE_PLUGIN_ROOT}/scripts/companion.mjs" delegate ... and returns that command's stdout as-is.
• Return the companion stdout verbatim to the user.
• Do not paraphrase, summarize, rewrite, or add commentary before or after it.
• Do not ask the subagent to inspect files, monitor progress, poll /anymodel:status, fetch /anymodel:result, call /anymodel:cancel, summarize output, or do follow-up work of its own.
• If the helper reports that the engine is missing or unauthenticated, stop and tell the user to run /anymodel:setup.
• If the user did not supply a request, ask what the engine should investigate or fix.