The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation script fetches the ocr CLI tool and its dependencies from official Alibaba GitHub repositories and the npm registry, which are recognized as trusted services.
[REMOTE_CODE_EXECUTION]: The skill uses a curl | sh pattern to execute an official installer script provided by the Alibaba Open Code Review project.
[COMMAND_EXECUTION]: The skill executes local shell commands to run the code review tool and includes sudo operations during the build-from-source installation method to place binaries in system-wide directories.
[PROMPT_INJECTION]: As the skill's primary function is to ingest and analyze untrusted code (including diffs and whole files), it is inherently exposed to indirect prompt injection if the processed codebase contains malicious instructions designed to manipulate the LLM's feedback.
[DATA_EXFILTRATION]: The skill manages LLM provider credentials through environment variables and local configuration. It adheres to security best practices by explicitly instructing the agent to prompt the user rather than hardcoding API keys.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface Analysis: 1. Ingestion points: Reads Git diffs and local source files via ocr review and ocr scan. 2. Boundary markers: No explicit boundary markers or 'ignore' instructions are added to the code data before it is passed to the tool. 3. Capability inventory: The skill can execute shell commands via Bash and write to the filesystem. 4. Sanitization: No specific filtering or escaping of the ingested code content is performed by the wrapper scripts.

open-code-review