[andersalm]

Add a standalone GitHub build for the pinned llama.cpp CPU engine on Linux ARM64. It packages the required libraries and licenses, checks architecture and library compatibility, then loads a hash-verified SmolLM2 fixture and requires a text reply. The workflow uses an Ubuntu 22.04 container on the native ARM runner and uploads a review artifact.

Validation: build 36501810782 passed on commit 4726b0c6. Local shell/YAML checks, validator failure fixtures, formatting and Home entropy checks passed. The downloaded archive and engine checksums match the build receipts.

This two-file change starts from upstream/0.7.1-dev at 6c61c990. Runtime installer integration and a test on the actual Jetson remain the next steps. Release publication requires separate approval.

Part of #97.

[andersalm]

Replaced for ongoing CI input by PR #200. #133, #136 and #153 now inherit its same pinned ARM recipe, current-run artifact and SHA-256 check; source/refusal tests and one development review pass. #200’s develop merge still requires current CI and the operator’s different-family adversarial PASS.

The original recipe/history and prepared Jetson hardware inputs remain preserved. This closes the standalone proof PR as replaced and leaves #97’s installed hardware acceptance open.