The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks because its primary workflow involves the AI agent processing untrusted external data using Python scripts. A malicious user could embed instructions within interview transcripts, feature descriptions, or configuration JSONs to influence the agent's behavior after it analyzes the script outputs.
Ingestion points: Untrusted data enters the context via customer_interview_analyzer.py (text transcripts), rice_prioritizer.py (CSV feature lists), landing_page_scaffolder.py (JSON configuration), and project_bootstrapper.py (JSON configuration).
Boundary markers: Absent. The scripts do not use specific delimiters or warnings to isolate user-supplied content from the tool's generated analysis.
Capability inventory: The agent can use the included scripts to write generated code and project structures directly to the file system.
Sanitization: The landing_page_scaffolder.py script uses HTML escaping for its HTML output mode, but most other interpolation points in code templates (e.g., TSX components) lack specific filtering for injection patterns.
[COMMAND_EXECUTION]: Several automation scripts possess extensive file system access rights required for their legitimate scaffolding functions.
project_bootstrapper.py creates multiple directories and writes boilerplate files—including package.json, Dockerfile, and .env.example—to a user-provided output path using os.makedirs and open().write().
landing_page_scaffolder.py writes entire React/Next.js components or HTML pages to a file path provided as an argument, allowing the agent to persist generated code to the user's workspace.

product-skills