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
This release turns caretaker from partial triage into a deployable loop: ingestion can re-triage on issue comments, workable specs publish to a ready-for-code Pub/Sub topic, locks clear on NEEDS_HUMAN, and capacity exhaustion is a terminal error. Firestore now tracks error and pr_number. The new triage/eval Cloud Run entrypoints, judge runner, and local golden-issue sync tools let teams measure triage quality before scaling. Use the GCP deployment script and eval framework so automated triage is reproducible, recoverable, and gated by golden cases rather than ad-hoc prompts.
Proposed actions
- Upgrade to Gemini CLI v0.56.0-nightly.20260808.gcf22ac7e8 and run the release’s GCP deployment script to deploy caretaker agent services (triage Cloud Run job + eval runner entrypoint).
- Update caretaker Firestore to the new schema fields error and pr_number; confirm locks clear on NEEDS_HUMAN and that capacity-exhaustion failures are treated as terminal errors in your job retries.
- Enable ingestion issue-comment handling so new comments trigger the re-triage workflow, and subscribe your coding pipeline to the ready-for-code Pub/Sub topic for workable spec events.
- Collect a local golden issue set with the caretaker-evals golden-issue collection tool, sync it to Firestore, and run the triage judge runner via the Cloud Run eval job entrypoint after each prompt/orchestrator change.
- After prompt hill-climbing changes in caretaker-triage, re-run the triage evaluation framework on the golden set and only promote orchestrator updates that improve judge scores without increasing NEEDS_HUMAN or terminal error rates.
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 This release turns caretaker from partial triage into a deployable loop: ingestion can re-triage on issue comments, workable specs publish to a ready-for-code Pub/Sub topic, locks clear on NEEDS_HUMAN, and capacity exhaustion is a terminal error. Firestore now tracks error and pr_number. The new triage/eval Cloud Run entrypoints, judge runner, and local golden-issue sync tools let teams measure triage quality before scaling. Use the GCP deployment script and eval framework so automated triage is reproducible, recoverable, and gated by golden cases rather than ad-hoc prompts. ## Suggested actions 1. Upgrade to Gemini CLI v0.56.0-nightly.20260808.gcf22ac7e8 and run the release’s GCP deployment script to deploy caretaker agent services (triage Cloud Run job + eval runner entrypoint). 2. Update caretaker Firestore to the new schema fields error and pr_number; confirm locks clear on NEEDS_HUMAN and that capacity-exhaustion failures are treated as terminal errors in your job retries. 3. Enable ingestion issue-comment handling so new comments trigger the re-triage workflow, and subscribe your coding pipeline to the ready-for-code Pub/Sub topic for workable spec events. 4. Collect a local golden issue set with the caretaker-evals golden-issue collection tool, sync it to Firestore, and run the triage judge runner via the Cloud Run eval job entrypoint after each prompt/orchestrator change. 5. After prompt hill-climbing changes in caretaker-triage, re-run the triage evaluation framework on the golden set and only promote orchestrator updates that improve judge scores without increasing NEEDS_HUMAN or terminal error rates. Please: 1. Map the change to concrete pilot steps 2. Flag security / permission implications 3. Keep the pilot reversible
Your loop
This browser · no sign-in · not shared as “you”Community signal · 0 approved · 0 deferred · 0 ignored
Anonymous public tally (one vote per visitor; changing choice updates your vote). Not moderated product truth.