The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including ./ai-docs/PRD.md, user-provided feature descriptions, and various reference files like schemas and API contracts. This creates a surface for indirect prompt injection where malicious instructions in documents could attempt to influence agent behavior.
Ingestion points: Data enters the context via the ./ai-docs/PRD.md file, the $ARGUMENTS variable in User Input Mode, and files found within the ./ai-docs/references/ directory.
Boundary markers: The skill relies on structured templates located at references/spec-template.md and references/features-template.md to delimit generated content.
Capability inventory: The skill has the capability to write files to the filesystem (Write tool) and execute basic directory and file discovery commands (Bash tool).
Sanitization: Content is mapped to specific sections of a Markdown template, which provides a level of structural validation but does not explicitly sanitize the text for embedded LLM instructions.
[COMMAND_EXECUTION]: The skill utilizes Bash for standard operational tasks including directory creation (mkdir -p) and locating reference materials (find). These operations are restricted to the local project structure and do not involve execution of untrusted remote code.

feature