The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads external project files and user-provided arguments to construct a design brief.
Ingestion points: Processes content from CLAUDE.md, BRIEF.md, the reference/ directory, and $ARGUMENTS.
Boundary markers: Lacks explicit delimiters or instructions to ignore embedded directives in the sourced files.
Capability inventory: Performs file system reads, directory listings (ls), and writes a new BRIEF.md file.
Sanitization: No sanitization of ingested content is performed prior to inclusion in the generated brief.
[COMMAND_EXECUTION]: The skill executes a directory listing command to verify the presence of project configuration.
Evidence: Utilizes ls ${CLAUDE_SKILL_DIR}/../../../tokens to detect the project environment.
Context: The path traversal is used to locate design tokens relative to the skill's installation path within the project.

grill-me