av/harbor v0.5.19 on GitHub

harbor up / restart: the port pre-check no longer treats Harbor's own container as a conflict when its port is published on a specific address such as 127.0.0.1 (#264 by @AmirF194, fixes #261).
harbor up: no longer fails with container ... exited (0) when a one-shot init sidecar such as speaches-init finishes before Compose checks it. A sidecar that exits with an error still fails up (#298 by @brandon-nuttall).
Hermes: starts again with current nousresearch/hermes-agent images, which no longer ship PyYAML, when a backend (Ollama, llama.cpp, etc.) is connected (#289 by @brandon-nuttall).
AutoGPT: the service now runs the current AutoGPT Platform image instead of the retired Auto-GPT Classic, with Ollama and Traefik integrations (#256 by @ntindle).
Boost: Anthropic-style /v1/messages requests to OpenAI gpt-5 and o-series models now send max_completion_tokens instead of the rejected max_tokens (#301, fixes #299, reported by @swf1996120).

Also includes the Boost security fix from v0.5.18.

Full Changelog: v0.5.18...v0.5.19