rescue-stream

Use this when the Gemini run is long, open-ended, or write-capable and you want Claude to react to interim progress (file edits, tool calls, partial findings) instead of waiting for one final blob.

Use /gemini:rescue (not this command) when you only need Gemini's final answer as a single artifact.

Raw user request: $ARGUMENTS

Operating contract:

• Run a Bash call to launch the streamed background job:

  node "${CLAUDE_PLUGIN_ROOT}/scripts/gemini-companion.mjs" task --stream "$ARGUMENTS"
  

  The companion prints a JSON-ish payload (jobId, eventsFile, logFile).
• Capture the jobId and eventsFile from that output.
• Tail the job with node "${CLAUDE_PLUGIN_ROOT}/scripts/gemini-companion.mjs" tail "$jobId" (run_in_background: true) and read new lines via BashOutput periodically. Do not use raw tail -F — it leaks. The companion's tail exits on terminal job status.
• Each event line is JSON-line (stream-json). Common type values include init, message, tool_use, error. Treat lines that fail to parse as raw progress text.
• The companion's tail exits on its own when the job is completed / failed / cancelled. If it's still running after 30 s with no new lines, surface a stall.
• After completion, fetch the final response with:

  node "${CLAUDE_PLUGIN_ROOT}/scripts/gemini-companion.mjs" result <jobId>
  

  and return that verbatim, exactly like /gemini:rescue.

Argument handling:

• Preserve the user's natural-language task text exactly.
• --model, --read-only, --yolo, --sandbox are runtime flags — pass them through.
• Do not strip --stream; the helper requires it.
• Default to --read-only when the user asked for review/diagnosis/research with no explicit request to apply fixes.

Don'ts:

• Do not paraphrase or summarize Gemini's events. Surface them verbatim.
• Do not fall back to /gemini:rescue if streaming fails — surface the error and stop.
• Do not auto-apply fixes from a streamed review. Same rule as /gemini:rescue.