The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a data processing pipeline that ingests multiple versioned specification files (V*.md) to generate a consolidated CURRENT.md file. This file is specifically designed to be used as a context prompt for external AI design tools (e.g., v0, Figma Make). This architecture presents an indirect prompt injection surface where malicious instructions in a specification file could influence downstream agents.
Ingestion points: The skill reads versioned Markdown files in the products/specs/<product-line>/ directory and scans project files such as src/pages/, src/routes/, and CLAUDE.md during initialization.
Boundary markers: The skill does not implement explicit delimiters or "ignore embedded instructions" warnings when merging content from versioned files into the consolidated CURRENT.md snapshot.
Capability inventory: The skill is granted Read, Write, Edit, and Bash capabilities, which are used to traverse the filesystem, read source code for context, and write synthesized documentation.
Sanitization: There is no evidence of content sanitization or instruction filtering for the Markdown text being aggregated. If a versioned file were modified to include malicious prompt instructions, these would be included in the consolidated output.

product-spec