The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow begins with the parsing of user-provided 'design documents' to generate code. This creates a surface for indirect prompt injection if a document contains malicious instructions intended to influence the agent's code generation or command execution behavior.
Ingestion points: The workflow in SKILL.md starts with '设计文档解析' (Design document parsing).
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded prompts within the design documents.
Capability inventory: The agent has the capability to write multiple C++ and CMake files to the local file system and execute shell commands (bash build.sh).
Sanitization: No explicit sanitization or validation of the design document content is mentioned before it is used for code generation.
[COMMAND_EXECUTION]: The skill instructs the agent to execute local shell commands to generate operator skeletons and run tests.
Evidence: SKILL.md and references/example/example-rules.md contain instructions to run bash build.sh --genop=..., bash build.sh --pkg ..., and bash build.sh --run_example ... within the user's project directory.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading external library code if it is missing from the project environment.
Evidence: references/op_kernel/kernel-rules.md specifies the command git clone https://gitcode.com/cann/catlass.git catlass to fetch the necessary Catlass source code from GitCode, a well-known developer platform.

catlass-operator-code-gen