The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external .R scripts. If a script contains malicious instructions, the agent or the sub-agent (r-reviewer) could be influenced to deviate from the protocol.
Ingestion points: The content of .R files identified via $ARGUMENTS or directory globbing in scripts/R/ and Figures/*/.
Boundary markers: No explicit delimiters or instructions to ignore embedded prompts are provided when passing the script content to the r-reviewer agent.
Capability inventory: The skill uses Read, Grep, Glob, Write, Agent, and Task tools. While Edit is disallowed, the Write tool could potentially be abused to write to unintended locations if the agent is compromised by instructions within the scripts being reviewed.
Sanitization: No sanitization or filtering of the input file content is performed before processing.

review-r