The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches and installs the open-code-review package from the official @alibaba-group organization repository on the npm registry.
[COMMAND_EXECUTION]: Executes the ocr command-line utility for reviewing code changes, performing audits, and managing historical review sessions.
[COMMAND_EXECUTION]: Executes a local verification script, check-config.sh, which replays configuration settings against a local mock server to validate the transmission of thinking-control parameters.
[DYNAMIC_EXECUTION]: Generates and executes Python logic embedded within the check-config.sh script to patch local JSON configuration files and inspect captured request logs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted source code and diff data from the local workspace to be processed by a remote LLM gateway, creating a vulnerability surface where code-embedded instructions could influence agent behavior.
Ingestion points: Code review targets include staged/unstaged changes, commit objects, and full directory scans initiated via the ocr review and ocr scan commands in SKILL.md.
Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" directives to isolate the code content from the analysis prompts.
Capability inventory: The skill can execute arbitrary shell commands via the ocr tool and local scripts, and it maintains persistent state in the ~/.opencodereview/ directory.
Sanitization: There is no evidence of sanitization or filtering of workspace content before it is transmitted to the LLM provider for review.

ocr