The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes App Store metadata retrieved from external sources, creating a vulnerability surface for indirect prompt injection.\n
Ingestion points: Metadata is retrieved from App Store Connect using asc localizations download (Step 2) and subsequently interpolated into the "LLM Translation Prompt Template" (Step 3) for processing by the model.\n
Boundary markers: The translation prompt uses triple quotes (""") for longer fields like description and whatsNew, which provides minimal structural separation but lacks specific instructions for the agent to ignore potentially malicious commands embedded within the text.\n
Capability inventory: The skill is configured to execute shell commands via the asc CLI, including administrative actions such as asc localizations upload and asc apps info edit.\n
Sanitization: There is no evidence of input validation, character filtering, or sanitization applied to the metadata before it is presented to the LLM for translation.\n- [COMMAND_EXECUTION]: The skill relies extensively on the asc CLI to perform its functions.\n
The workflow involves executing multiple shell commands to list apps, versions, and metadata, as well as to upload translated content.\n
The reliance on a command-line interface for tasks involving external data ingestion increases the risk that an indirect injection attack could be translated into unauthorized command execution.

asc-localize-metadata