akg-agents

To automate AKG kernel code generation, you must provide a mandatory `~/.akg/settings.json` file and run `run_workflow.py --workflow kernelgen` in the foreground. The workflow enforces the correct AKG repository branch, initializes the environment, extracts tasks, and executes the generation deterministically.

Environment setup for AKG operator workflow execution forces a full initialization when `~/.akg/check_env.md` is absent. The process reads the required `akg-env-setup` skill to bootstrap the environment before proceeding with any kernel generation or verification tasks.

Before running the AKG kernelgen workflow, you need to prepare a `~/.akg/settings.json` file specifying your selected framework, backend, dsl, and architecture. This settings file is mandatory for the workflow to execute correctly and generate the kernel code.

No, AKG kernel generation must run in the foreground. The workflow requires explicit execution of `run_workflow.py --workflow kernelgen` to ensure deterministic orchestration and truthfully report any command failures without silently switching methods.

Your AKG workflow fails without `settings.json` because it is a strict requirement located at `~/.akg/settings.json`. The workflow relies on this file to obtain user-selected framework, backend, dsl, and arch configurations required for task extraction and kernelgen code generation.

AKG operator authoring and performance tuning involves cloning the AKG repository to a pinned branch, initializing the environment via `akg-env-setup`, extracting standard task files, and running the kernelgen workflow. This deterministic orchestration handles verification and backend implementation completion scenarios.