[Alex R (alexr17)]

Builds on #120 with opt-in conversation-based Loop Runtime autoscaling through GKE custom metrics, defaulting to 1–50 pods at 50% claimed capacity.
Applications expose loop_runtime_claimed_conversation_utilization (0–100) on their runtime HTTP metrics endpoint; missing application support leaves the HPA unable to perform metric-driven scaling, with no image-version gate.
Helm leaves replica ownership to the HPA while enabled, applies a five-minute downscale window, and restores fixed replicas when disabled.
Sandbox-worker replicas remain separately configured and must provide enough capacity for the runtime limit.

Input	Behavior
Google Autopilot or Standard, runtime and autoscaling enabled, AutoscalingMetric API present	Render the HPA and metric source; omit Deployment replicas.
Autoscaling disabled, missing, or null	Preserve fixed replicas, including explicit zero and the null/empty fallback to one.
Runtime disabled or missing from reused values	Omit runtime autoscaling resources.
Partial autoscaling map on reused values; missing or null settings	Fill the documented autoscaling defaults.
Disabling autoscaling after enabling it	Remove the HPA and metric source; restore the configured fixed count.
Enabling on an existing Deployment	Removing the replica field can initially reset the count to one.
Empty, zero, negative, fractional, or reversed replica bounds	Reject invalid autoscaling bounds.
Empty, zero, negative, or above-100 utilization; empty or relative metrics path	Reject invalid enabled settings.
Missing AutoscalingMetric API	Fail with prerequisite guidance.
AWS or Azure	Runtime remains disabled by default; enabling it retains the existing Google-only validation.
Release tags with or without v, full/short/numeric SHAs, prerelease and custom tags	Pass through unchanged, including images without the metrics endpoint.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-03T23:58:39.347699Z	85d7501	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review" or "Codex (@codex) security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 85d7501ff5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "Codex (@codex) address that feedback".

[chatgpt-codex-connector]

Cover the required autoscaling upgrade and empty-value matrix

The new suite does not exercise the helper's reuse-values/defaulting branches: every enabled case inherits the complete autoscaling defaults, and the null case disables the entire map rather than testing null settings while enabled. There are also no cases for reversed or fractional replica bounds, despite the commit explicitly promising to reject them. Add cases that remove the retained autoscaling map before applying partial overrides, supply null individual settings, and exercise the distinct invalid-bound branches. These are explicitly required release-blocking checks under AGENTS.md, not optional coverage improvements.

AGENTS.md reference: AGENTS.md:L14-L17

Useful? React with 👍 / 👎.