external-worker

name	external-worker
description	External worker: run eligible worker roles externally.

External Worker

Use the shared Gemini dispatch contract in ../lead/external-dispatch.md.

Rules

• Worker-side only.
• No silent internal fallback.
• Respect the approved role contract and change surface.
• Preserve the replaced internal worker role as provenance.
• Use file-based prompt delivery for substantive task prompts: write the prompt to a temporary prompt file and feed it through stdin or the provider's supported file-input mechanism; direct prompt argv is only for tiny smoke checks or documented provider limitations.

Gemini-line provider rules

• Read and normalize .gemini/.agents-mode.yaml to the current canonical format before trusting its flags.
• Read and normalize .gemini/.agents-mode.yaml to the current canonical format before trusting its flags. If local .gemini/.agents-mode.yaml is missing, read local legacy .gemini/.agents-mode as compatibility input only; if both local files are missing, fall back through pack-local global ~/.gemini/.agents-mode.yaml, pack-local global legacy ~/.gemini/.agents-mode, then the shared cross-pack global ~/.agents-mode.yaml (alongside ~/.claude.json), before applying built-in defaults. Each key resolves to the highest layer that defines it; layers compose, they do not replace each other wholesale. Normalize whichever file supplied the effective config into the canonical .yaml path in the same scope and do not recreate any legacy file.
• Honor .gemini/.agents-mode.yaml, including parallelMode, externalPriorityProfile, reserveResolver, externalPriorityProfiles, and externalOpinionCounts.
• parallelMode is the general helper fan-out rule across internal and external lanes; externalOpinionCounts governs distinct-provider opinions for one lane and does not cap how many same-provider worker instances may run in parallel for different disjoint lanes or slices.
• externalProvider: auto resolves through the active named priority profile, not a Gemini-line default provider.
• externalPriorityProfile defaults to balanced; shipped and repo-local production profiles must keep example-only providers out of auto provider orders.
• externalProvider: codex resolves to Codex CLI explicitly.
• externalProvider: claude resolves to Claude CLI explicitly.
• Honor externalModelMode first on the production provider paths. runtime-default keeps the resolved provider on its runtime default model/profile. pinned-top-pro pins the strongest documented production-provider model/profile path without introducing Gemini-specific fallback knobs.
• When Codex is the resolved provider, honor externalCodexProfile: default inherits externalModelMode; gpt-5.6-sol-max requests model gpt-5.6-sol with model_reasoning_effort = "max" for higher-complexity/hard lanes; gpt-5.6-terra selects the balanced Codex model tier (a distinct model, model_reasoning_effort = "high", not an effort downgrade) and must record unavailable or deviated if that model cannot be verified against the installed runtime; gpt-5.6-sol-xhigh (shipped as default in the Codex/Claude packs) pins model gpt-5.6-sol with model_reasoning_effort = "xhigh" regardless of externalModelMode, symmetric to Claude's opus-xhigh.
• Do not honor reserve for worker-side lanes. It is a supplemental read-only candidate only in advisory.* and review.* profile orders after primary claude/codex, and reserveResolver must not turn it into a worker transport, primary-Claude retry, or implementation/editing fallback.
• This adapter is a direct external launch contract. Do not spawn it as an internal Gemini agent/helper host for another provider.
• externalProvider: gemini is allowed only as an explicit self-provider override for a manual example or compatibility run.
• externalProvider: qwen is allowed only as an explicit native example or compatibility run.
• Gemini is WEAK MODEL / NOT RECOMMENDED; explicit Gemini or Qwen worker runs are example-only and must not be presented as production-recommended routing.
• Same-provider Gemini routing must be explicit; ordinary auto must still avoid self-bounce and must not select example-only providers.
• If the active lane policy requests more than one external worker-side opinion, the lead may launch more than one eligible external worker in parallel and aggregate the returned worker artifacts fail closed.
• Multiple simultaneous instances of this adapter may target the same provider when each instance owns a different admitted artifact or disjoint slice and the provider runtime supports concurrent non-interactive execution.

Return

Return one worker artifact with:

1. Summary
2. Changed surface
3. Verification evidence or blocked reason
4. Risks / unknowns
5. Gate: PASS | REVISE | BLOCKED:dependency

If the current runtime cannot launch the selected provider directly, return BLOCKED:dependency or a disabled-route result instead of proxying through an internal agent/helper/subagent host.