The Agent Skills Directory

Indirect Prompt Injection Surface: The skill transforms untrusted Design Thinking markdown artifacts into YAML slide definitions. This creates an attack surface where adversarial content embedded in source files could attempt to influence the agent's behavior during subsequent processing by the PowerPoint skill pipeline.
Ingestion points: Markdown artifacts are read from the directory specified by the --canonical-dir flag in scripts/generate_cards.py (defaulting to .copilot-tracking/dt/<project-slug>/canonical).
Boundary markers: The skill does not currently implement explicit prompt delimiters or "ignore embedded instructions" warnings in its emitted YAML output, relying instead on the consumer skill's parsing logic.
Capability inventory: The skill's primary capabilities are limited to local file reading and writing. However, the generated output is used by the powerpoint skill, which possesses broader operational capabilities.
Sanitization: The yaml_escape function in scripts/generate_cards.py provides effective sanitization by escaping backslashes, double quotes, and newlines, ensuring that external content remains confined within its intended YAML scalars.
Command Execution: The documentation provides instructions for running Python scripts (generate_cards.py) and build tools (Invoke-PptxPipeline.ps1).
Context: These operations are performed locally by the operator as part of the intended development and rendering workflow.
External Downloads: The skill documentation recommends installing the uv package manager via well-known system package managers such as winget or Homebrew.
Context: This follows established security practices for software installation and dependency management.

customer-card-render