Deployed the AutoCoder autonomous self-improvement control plane and the first hosted-application feedback producer in Bodylog.
gitea.app.lan/autocoder/autocoder-engine:sha-d9de1374baaegitea.app.lan/bodylog/bodylog.dev:sha-f178a0082d031/1 ready replicas.4082 and 4083 completed successfully.autocoder-config:
AUTOCODER_SELF_IMPROVEMENT_FEEDBACK_TOKEN to the existing autocoder-secrets Secret in both autocoder-engine and bodylog. The value is not stored in Git or Wiki.js.bodylog-app-config.autocoder_self_improvement_feedback, autocoder_self_improvement_cycles, their uniqueness indexes, and per-worker token-count columns.GET /api/self-improvement/status returned enabled state, 30 percent reserve, 10 percent allowance, pay-as-you-go disabled, and capacity_available=false.app=bodylog policy identity crossed both NetworkPolicies and persisted a Bodylog deployment-success event. The validation pod was deleted afterward.self-review-* worker rows were backed up to /home/ax_julian/temp/autocoder-legacy-self-review-queued-2026-07-23.jsonl and then cancelled. Backup SHA-256: 76abfe5ac38d3e15ea9cad669b6b7bffcbf59575a7e9e8fdb8866dcea8b07a68. Zero legacy rows remained queued.Initial builds were delayed when the node entered ephemeral-storage pressure and evicted an Actions runner. The managed containerd image prune could not schedule while the disk-pressure taint was active, so the same crictl rmi --prune operation was run directly. An additional host Docker build-cache prune limited to cache older than seven days reclaimed 14.14 GB. Disk pressure cleared, available root space recovered to about 58 GB, the runner recovered, and both repository workflows then completed successfully.
Revert the relevant repository commits and deploy through each repository's normal main workflow. Do not delete persisted feedback or cycle audit rows as part of routine rollback. Secret removal or database cleanup requires a separate, explicitly scoped operation.