The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several shell commands to manage the code review process, including repository interaction via git and gh, and processing via external analysis tools such as duplicatecode and pgdb.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted code diffs and pull request descriptions, which could contain instructions designed to influence the behavior of the orchestrator or its subagents.
Ingestion points: Filesystem data retrieved through git diff and gh pr diff commands.
Boundary markers: The diff content is saved to files and passed to subagents, but the instructions do not include specific delimiters or sandboxing directions to prevent the model from following instructions embedded within the analyzed code.
Capability inventory: The orchestrator and subagents have capabilities for filesystem access, shell command execution, and interaction with other API-based tools.
Sanitization: There are no explicit sanitization or filtering steps applied to the diff content before it enters the agent's context.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing and updating external tools, specifically the duplicatecode utility from the bmsuisse GitHub repository. While these are vendor-provided resources, they represent the execution of code downloaded from external sources during the developer workflow.

bms-code-review