The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the CodeArts codecheck engine and custom clang-tidy binaries from Huawei's Object Storage Service (OBS) during initialization.
Evidence: scripts/setup.sh uses curl to download zip archives from https://codeheck-chajian-bj4-obs.obs.cn-north-4.myhuaweicloud.com/ into the ~/.codecheck-tools/ directory.
[COMMAND_EXECUTION]: Executes the downloaded analysis engine and various system-installed linters to evaluate source code files.
Evidence: scripts/check.sh invokes java -jar for the CodeArts engine and runs shell commands for pylint, flake8, shellcheck, and gn format based on file extensions.
[PROMPT_INJECTION]: The skill processes untrusted code files which could theoretically contain malicious instructions targeting the agent via Indirect Prompt Injection.
Ingestion points: Processes local source files including .cpp, .py, .sh, and .gn files within the repository.
Boundary markers: The agent is instructed to parse structured, pipe-delimited output from the shell scripts rather than raw linter output.
Capability inventory: The agent has access to the Edit and Write tools to perform auto-fixes, and shell execution via the skill's scripts.
Sanitization: Scripts like check_python.sh and check_shell.sh include sed patterns to remove delimiter characters (|) from linter messages, ensuring the agent receives correctly formatted data and preventing interpretation of injected control characters.

oh-precommit-codecheck