The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx repomix to aggregate the codebase into a single XML file. This involves fetching and running the repomix package from the NPM registry at runtime.
[COMMAND_EXECUTION]: The skill executes several shell commands, including git rev-parse, mkdir, find, and npx. Notably, it implements a security check using a case statement to verify that the target review path does not traverse outside the project root, mitigating potential directory traversal risks.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest the entire codebase into the context of multiple subagents. This creates a surface for indirect prompt injection if an attacker places malicious instructions within the source code or documentation of the repository being reviewed.
Ingestion points: The npx repomix command in Step 3 reads all files within the target path into a single repomix.xml file.
Boundary markers: The instructions for the 9 subagents do not include specific delimiters or warnings to ignore instructions found within the processed code data.
Capability inventory: The orchestration process includes shell command execution, file system modifications (mkdir/find), and file writing via subagents to the .llmtmp/ directory.
Sanitization: There is no evidence of filtering or sanitizing the codebase content before it is passed to the reviewer agents.

review-full