The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes information from local design documents to generate executable Python code, creating a potential vector for indirect prompt injection if those source files contain malicious instructions.
Ingestion points: Reads content from SPEC.md, DESIGN.md, and {op}_golden.py as specified in SKILL.md under the '所需输入' section.
Boundary markers: None identified in the prompt templates to distinguish between documentation data and code logic.
Capability inventory: The skill writes Python scripts to the filesystem and facilitates execution through environment setup and test instructions.
Sanitization: No explicit content validation or escaping is observed during the interpolation of design data into the code templates.
[DYNAMIC_CONTEXT_INJECTION]: The skill uses shell command execution at load time to discover hardware status and initialize environment variables.
Evidence: Execution of bash scripts/list_idle_chip_ids.sh within SKILL.md to populate the TILE_FWK_DEVICE_ID environment variable.
[EXTERNAL_DOWNLOADS]: The environment preparation script fetches source code from an external repository to initialize the development environment.
Evidence: git clone https://gitcode.com/cann/pto-isa.git in scripts/environment_prepare.sh.
Context: The download originates from a recognized development platform (GitCode) and is consistent with the skill's stated purpose of preparing the PyPTO development environment.
[COMMAND_EXECUTION]: The skill relies on system utilities and shell scripts for environment verification, hardware discovery, and executing generated tests.
Evidence: Subprocess execution of npu-smi, awk, python3, and git as described in the provided scripts and setup instructions.

external-cannbot-ops-pypto-op-develop