The Agent Skills Directory

[COMMAND_EXECUTION]: The router.sh script is vulnerable to command injection through unsafe shell expansion. It uses unquoted or unsafely expanded positional parameters ($*) inside double-quoted echo statements, which allows for the execution of command substitutions (e.g., $(...) or backticks) if they are present in the arguments.
File: router.sh
Evidence: echo "Target: $", echo "Libraries: $", echo "Source: $", echo "Goal: $"
[PROMPT_INJECTION]: The skill design creates an indirect prompt injection surface as it is intended to process external documentation and harvest content from files.
Ingestion points: fetch operation (retrieving external library documentation from well-known sources like Drizzle ORM or Next.js) and harvest operation (processing summary files).
Boundary markers: None identified in the provided script or documentation.
Capability inventory: File system operations (claims to cleanup, organize, and compress) and shell execution via router.sh.
Sanitization: The provided shell script contains no sanitization or escaping logic for handled data.
[NO_CODE]: The router.sh script is a placeholder that only outputs text to the console. It does not implement the functional logic for the advertised operations such as file discovery, fetching, harvesting, or cleanup.

context-manager