Deploy caretaker triage Cloud Run jobs with comment re-triage and eval runner
Gemini CLI nightly adds caretaker triage Cloud Run workflow, issue-comment re-triage, ready-for-code Pub/Sub specs, Firestore error/pr_number fields, golden-issue eval tools, and a GCP deploy script—wire these into your issue-to-PR agent loop.Why this loop
Gemini CLI v0.56.0-nightly.20260808.gcf22ac7e8 ships the caretaker issue-to-PR path as deployable jobs, not just prompts. Triage is a Cloud Run job; ingestion now handles issue comments and re-triage; a workable spec is published to the ready-for-code Pub/Sub topic. Firestore gains error and pr_number. Capacity Exhaustion is a terminal error (do not retry). NEEDS_HUMAN clears the lock. caretaker-evals adds a triage eval framework, judge runner, local golden-issue collection, Firestore sync, and a Cloud Run eval-runner entrypoint, plus a GCP deploy script for caretaker services. caretaker-triage also updates prompt hill-climbing and the orchestrator. If you already run this loop, you must deploy the new jobs, migrate schema, and hook comment re-triage plus the Pub/Sub handoff or comments stall, exhausted capacity retries, and specs never reach code.
Proposed actions
- Check out google-gemini/gemini-cli at tag v0.56.0-nightly.20260808.gcf22ac7e8 and run the new GCP deployment script for caretaker agent services so the triage Cloud Run job workflow and the caretaker-evals Cloud Run job entrypoint for the eval runner are deployed in the same project as Firestore and Pub/Sub.
- Migrate caretaker Firestore documents to add error and pr_number fields; classify Capacity Exhaustion as a terminal error with no retry; on NEEDS_HUMAN, clear the lock so a human can take the issue.
- Point ingestion at GitHub issue comments so each comment enqueues the re-triage workflow (updated caretaker-triage hill-climbing prompts and orchestrator). When triage emits a workable spec, publish that event to the ready-for-code Pub/Sub topic for the code agent.
- Stand up caretaker-evals: collect local golden issues, sync them to Firestore with the new tools, then run the triage evaluation framework and judge runner through the Cloud Run eval-runner entrypoint against the updated orchestrator.
Agent prompt
Paste into your agent or query via MCP (get_agent_prompt) — free, no extra AI costGoogle Antigravity / agent task
DevAgentRadar → Google Antigravity
Goal: turn this release signal into a safe pilot plan for my stack.
Context
Assistant: Gemini CLI Proposal: Deploy caretaker triage Cloud Run jobs with comment re-triage and eval runner Summary: Gemini CLI nightly adds caretaker triage Cloud Run workflow, issue-comment re-triage, ready-for-code Pub/Sub specs, Firestore error/pr_number fields, golden-issue eval tools, and a GCP deploy script—wire these into your issue-to-PR agent loop. Primary source: https://github.com/google-gemini/gemini-cli/releases/tag/v0.56.0-nightly.20260808.gcf22ac7e8
Why it matters
Gemini CLI v0.56.0-nightly.20260808.gcf22ac7e8 ships the caretaker issue-to-PR path as deployable jobs, not just prompts. Triage is a Cloud Run job; ingestion now handles issue comments and re-triage; a workable spec is published to the ready-for-code Pub/Sub topic. Firestore gains error and pr_number. Capacity Exhaustion is a terminal error (do not retry). NEEDS_HUMAN clears the lock. caretaker-evals adds a triage eval framework, judge runner, local golden-issue collection, Firestore sync, and a Cloud Run eval-runner entrypoint, plus a GCP deploy script for caretaker services. caretaker-triage also updates prompt hill-climbing and the orchestrator. If you already run this loop, you must deploy the new jobs, migrate schema, and hook comment re-triage plus the Pub/Sub handoff or comments stall, exhausted capacity retries, and specs never reach code.
Suggested actions
- Check out google-gemini/gemini-cli at tag v0.56.0-nightly.20260808.gcf22ac7e8 and run the new GCP deployment script for caretaker agent services so the triage Cloud Run job workflow and the caretaker-evals Cloud Run job entrypoint for the eval runner are deployed in the same project as Firestore and Pub/Sub.
- Migrate caretaker Firestore documents to add error and pr_number fields; classify Capacity Exhaustion as a terminal error with no retry; on NEEDS_HUMAN, clear the lock so a human can take the issue.
- Point ingestion at GitHub issue comments so each comment enqueues the re-triage workflow (updated caretaker-triage hill-climbing prompts and orchestrator). When triage emits a workable spec, publish that event to the ready-for-code Pub/Sub topic for the code agent.
- Stand up caretaker-evals: collect local golden issues, sync them to Firestore with the new tools, then run the triage evaluation framework and judge runner through the Cloud Run eval-runner entrypoint against the updated orchestrator.
After you finish
Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.
Tell the human: open https://devagentradar.com/proposals/gemini-cli-v0-56-0-nightly-20260808-gcf22ac7e8-deploy-caretaker-triage-c and mark Applied, Skipped, or Failed. Proposal id: d56a5fdf-3f80-40a7-b473-e41c56ee3564
Please:
- Map the change to concrete pilot steps
- Flag security / permission implications
- Keep the pilot reversible
Your loop
This browser · no sign-in · not shared as “you”After you run the prompt
Only you can mark this. Agents cannot write your loop.
Your decision stays on this device. A public tally appears after a few votes.