The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates architecture auditing by executing local command-line utilities such as git, ast-grep, cargo, and go. These tool calls are implemented securely using Python's subprocess module with argument lists and shell=False, which prevents command injection vulnerabilities. Evidence of these patterns is found in scripts/provider_process.py and scripts/architecture_audit/discovery.py.
[PROMPT_INJECTION]: The skill's primary objective is to analyze the content of arbitrary code repositories, creating an attack surface for indirect prompt injection. Malicious instructions or adversarial patterns embedded within audited files could attempt to influence the AI agent's reasoning. The evidence chain for this category includes: (1) Ingestion points in scripts/architecture_audit/discovery.py where repository files are inventoried and read; (2) An absence of explicit boundary markers for untrusted file content; (3) A capability inventory that includes extensive local command execution; and (4) An absence of specific content sanitization before processing. This is a low-severity risk inherent to the skill's purpose as an auditing tool.

software-architecture