tasselszcx/ako-npu/skill

ops-easyasc-dsl

This is the user-facing entrypoint for the easyasc DSL to AscendC skill. The reusable workflow itself lives under agent/; this file only points you at it.

Before you read the archived runtime/docs or examples

Parts of this repository ship as archives to keep the delivered skill bundle small.

Run the initialization script once before reading runtime/doc content or running examples:

bash agent/scripts/init.sh

agent/scripts/init.sh restores:

• easyasc/ — the DSL Python package and codegen runtime
• doc/ and doc_cn/ — long-form English and Chinese documentation
• agent/example/ — curated kernel examples and manual demo programs

The script is idempotent; only missing trees are restored. Archives live under agent/assets/:

• agent/assets/ops-easyasc-dsl-runtime.tar.gz — the easyasc/, doc/, and doc_cn/ payload
• agent/assets/ops-easyasc-dsl-example.tar.gz — the agent/example/ payload

Where to go next

Start with the router:

• agent/ROUTER.md — progressive-disclosure routing layer

Preferred read order:

1. agent/ROUTER.md — pick the matching route
2. one playbook from agent/playbooks/
3. one focused reference from agent/references/
4. one source example from agent/example/kernels/ (after init.sh)

Top-level layout

• skill/ — this skill entrypoint
• agent/ — the callable easyasc DSL to AscendC workflow used by this skill
• agent/scripts/ — repository-maintenance scripts, including init.sh
• agent/assets/ — archived runtime/docs and example payloads
• agent/example/ — restored kernel examples and manual demos (on-demand)

Environment guidance

The delivered scripts prefer environment variables over fixed install paths. Set them to match your CANN installation, for example:

export ASCEND_HOME_PATH=/path/to/your/cann/install
# optional conda env name (example only, not required)
# conda activate torch210npu

Avoid hardcoding machine-specific absolute paths when editing delivered scripts.