The Agent Skills Directory

[COMMAND_EXECUTION]: Potential shell command injection in the workflow's final step.
Evidence: The skill executes $CLAUDE_PLUGIN_ROOT/scripts/project-status.sh <project-dir> where <project-dir> is a variable derived from user input or project discovery.
Risk: Lack of input validation or sanitization on the directory path could allow for arbitrary command execution if a path containing shell metacharacters is processed.
[PROMPT_INJECTION]: Surface for indirect prompt injection through external project files.
Ingestion points: portfolio.json, products/*.json, and features/*.json in the current working directory.
Boundary markers: No delimiters or warnings are used when processing content from these files.
Capability inventory: The skill utilizes Bash and Write tools, which could be abused if the agent follows instructions embedded in the project data.
Sanitization: Information from data fields (such as product names, feature purposes, and language settings) is interpolated directly into the agent's context and the generated diagram without validation.

portfolio-architecture