| name | gemini-sync-loop |
| description | Run a synchronous local Gemini loop for this repository that authors one task, executes it, triages the report, and repeats until stopped or quota is exhausted. Use when the user wants unattended Gemini progress while away from the keyboard. |
Gemini Sync Loop
Use this skill when the user explicitly wants a local watcher that keeps Gemini moving without manual "finished" messages between iterations.
What It Does
- Runs on the local machine in this repository.
- Calls
gemini in headless mode and waits synchronously for each invocation to finish.
- Uses the existing communication folders:
communication/CodexToGemini/command_waiting
communication/CodexToGemini/command_processing
communication/CodexToGemini/command_processed
communication/GeminiToCodex/report_waiting
communication/GeminiToCodex/report_completed
communication/GeminiToCodex/report_failed
- Honors:
communication/auto_agent_no_loop
communication/auto_agent_state.json
Safety Rules
- This loop is local only. It is not a hidden Codex background feature.
- It must not push.
- It must stop on Gemini quota/rate-limit errors.
- It must stop when
communication/auto_agent_no_loop exists.
- It should manage only work created after the loop starts, so stale backlog is not accidentally reprocessed.
Runner
- Main runner:
/Volumes/PoppoSSD2T/Projects/ClaudeCodeProjects/Legacy89DiskKit/scripts/gemini_sync_loop.py
Decision Order Per Iteration
- If a managed report is waiting, triage exactly one report.
- Else if a managed instruction is waiting, process exactly one instruction.
- Else author exactly one next conservative instruction.
- Stop on repeated no-op iterations, explicit stop marker, or quota exhaustion.
Output Expectations
- Update
communication/auto_agent_state.json
- Append logs to
communication/auto_agent_loop.log
- Keep each Gemini invocation single-purpose and synchronous