The Agent Skills Directory

REMOTE_CODE_EXECUTION (CRITICAL): Confirmed detection of piped remote execution via curl -fsSL https://bun.sh/install | bash. This executes unverified code from an external domain that is not on the trusted sources list.
EXTERNAL_DOWNLOADS (HIGH): The skill downloads and executes scripts from bun.sh. Since this domain is not a verified trusted source (e.g., official GitHub organizations like Google or Microsoft), the integrity of the downloaded content cannot be guaranteed.
PROMPT_INJECTION (LOW): scripts/lint-check.sh is vulnerable to indirect prompt injection (Category 8).
Ingestion points: The script reads raw tool input from stdin via input=$(cat) and parses it with jq to extract file_path.
Boundary markers: None. There are no delimiters or instructions to ignore embedded commands in the processed data.
Capability inventory: The script executes biome, eslint, and prettier on the provided file path.
Sanitization: While it uses jq -r and shell quoting, it lacks validation to prevent path traversal or ensure the target file is within an expected directory.
COMMAND_EXECUTION (LOW): scripts/test-runner.sh executes bun test, which runs code defined in the project's test files. This is a standard but high-privilege capability that could be exploited by malicious project files.

bun-expert