The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata from App Store Connect via the asc localizations download command and interpolates it into an LLM translation prompt. An attacker who controls the source metadata could theoretically attempt to influence the LLM's output or behavior.
Ingestion points: Metadata is downloaded to local .strings files via asc localizations download as described in Step 2 of the workflow.
Boundary markers: The LLM Translation Prompt Template uses triple-quote delimiters (""") to separate instructions from source text, which provides a layer of defense against accidental obedience to embedded instructions.
Capability inventory: The skill possesses the ability to execute asc CLI commands to modify app metadata on App Store Connect.
Sanitization: The skill includes a mandatory human-in-the-loop safety check in the Agent Behavior section: "Show the user translations before uploading — present a summary table of all fields × locales for approval. Do not push without confirmation."
[COMMAND_EXECUTION]: The skill relies on executing shell commands via the asc CLI tool to interact with App Store Connect services. This includes downloading, uploading, and listing app localizations and version information. This behavior is the primary intended function of the skill and uses explicit flags and IDs to ensure deterministic operations.

asc-localize-metadata