The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local Bash commands including grep, echo, and cut to automate the extraction of parameters and architecture details from files in the workspace/ directory. These commands are integral to the skill's purpose and are scoped to the local filesystem.\n- [SAFE]: Analysis of the automated scripts reveals that extracted data is filtered through a strict regular expression (\w+), which prevents potential command injection by ensuring variable names consist only of alphanumeric characters and underscores.\n- [SAFE]: Regarding indirect prompt injection: 1. Ingestion points include workspace/paper/main.tex and workspace/src/config.py. 2. Boundary markers are not explicitly used in the prompt logic. 3. Capabilities include file read access and local command execution. 4. Sanitization is implemented via regex validation of extracted keys before shell use.

code-audit