The Agent Skills Directory

[COMMAND_EXECUTION]: The skill explicitly instructs the agent to use Bash for file writing operations (cat << 'EOF' > file) as a fallback for the standard Write tool. Critically, it directs the agent to perform these actions autonomously and "silently" without seeking user permission, which reduces transparency and user oversight of shell-level operations.- [PROMPT_INJECTION]: The workflow is vulnerable to Indirect Prompt Injection. It ingests untrusted external data via the $ARGUMENTS parameter (typically a path to a narrative research report) and uses this content to drive the automated generation of LaTeX code, figures, and iterative improvements.
[PROMPT_INJECTION]: Ingestion points: The path or content of the narrative report provided in $ARGUMENTS (SKILL.md).- [PROMPT_INJECTION]: Boundary markers: None. The skill does not provide delimiters or instructions for the agent to ignore embedded commands or malicious directives within the input report.- [PROMPT_INJECTION]: Capability inventory: The skill has access to Bash(*), Write, Edit, and the ability to call other Agent and Skill components. This allows it to modify the filesystem and execute system commands.- [PROMPT_INJECTION]: Sanitization: None. The instructions direct the agent to parse the report for claims, evidence, and figure descriptions without validation or filtering, meaning malicious content in the report could influence the agent's behavior during the writing or improvement phases.

aris-paper-writing