The Agent Skills Directory

[COMMAND_EXECUTION]: The activation sequence in SKILL.md executes a local Python script (_bmad/scripts/resolve_customization.py) to resolve and merge configuration settings from customize.toml and other project-level override files.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it actively discovers and loads external markdown files (e.g., *brief*.md, *research*.md) into the agent's context to inform the PRD creation.
Ingestion points: Identified in steps-c/step-01-init.md, which searches for and loads files matching patterns like *brief*.md, *research*.md, and project-context.md from the project directory.
Boundary markers: The instructions do not define specific delimiters or instructions to the LLM to ignore potentially malicious directions embedded within those discovered documents.
Capability inventory: The skill has the capability to execute local scripts (via SKILL.md) and perform CSV data lookups (via step-02-discovery.md).
Sanitization: There is no explicit sanitization or filtering of the content loaded from the discovered external documents before they are used to influence the agent's output.

bmad-create-prd