The Agent Skills Directory

PROMPT_INJECTION (LOW): Indirect Prompt Injection surface detected via codebase markers.
Ingestion points: scripts/scan-codebase.sh extracts text from source files using grep for markers like AICODE-NOTE.
Boundary markers: Absent. The agent receives extracted text without delimiters or instructions to treat it as untrusted data.
Capability inventory: The skill is granted Bash (*) and Read permissions, creating a risk if the agent's logic is hijacked by injected instructions.
Sanitization: No sanitization is applied to the extracted comment text before it reaches the agent.
COMMAND_EXECUTION (SAFE): The skill runs a local script scripts/scan-codebase.sh. Analysis shows it is a well-behaved utility for gathering project metadata and does not perform network requests or unauthorized file modifications.

code-analyzer