The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses rg (ripgrep) or grep to bootstrap the analysis scope by searching the codebase for relevant files. The search patterns are dynamically constructed from keywords extracted from user requirements provided in $ARGUMENTS. If user-supplied input contains shell metacharacters (such as single quotes that could break out of the pattern wrapping), it could lead to unintended command execution.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes content from the local codebase and user requirements, which are then passed to sub-agents as instructional context.
Ingestion points: User-provided requirements in $ARGUMENTS, the contents of files identified during the codebase search, and user-specified prototype code paths.
Boundary markers: The instructions do not define explicit delimiters or instructions to sub-agents (like codebase-analyzer or ui-analyzer) to distinguish between system instructions and untrusted data from the codebase.
Capability inventory: The workflow involves shell command execution for file searching, spawning multiple specialized sub-agents via spawn_agent, reading files from the repository, and writing design artifacts to the docs/ directory.
Sanitization: The skill lacks explicit sanitization, validation, or filtering of the content ingested from the codebase before it is used to influence agent behavior.

recipe-front-design